Board

What is AI?

Technology263 insights0 followers

@technology
Calling a program “UNDERSTAND” or “GOAL” can smuggle the missing capability into its name and fool researchers.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
Reasoning models automate chain-of-thought prompting by generating thinking tokens, then feeding those reasoning traces back into the model.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
A model can imitate enough reasoning examples to produce a verifiable answer without learning a general reasoning algorithm.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
A reasoning trace can look like an auditable record of model thought while having no reliable connection to the reasoning that occurred.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
Large reasoning models may retrieve approximate answers from training data, landing closer to pattern matching than genuine reasoning.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
State-of-the-art reasoning models often work because surrounding conventional software guides and verifies outputs in agentic systems and theorem provers.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
A black-box AI is useful when outputs are verifiable, but research demands knowing whether it is right for the right reasons.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
Most reasoning models probably do not learn algorithms. They exploit larger example sets and increasingly clever reward signals.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
A reasoning model may be simply an LLM with specialized training, not a fundamentally new reasoning mechanism.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
A model’s hidden reasoning tokens can be arbitrary embeddings, foreign-language fragments, exclamations, or punctuation.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
OpenAI has not publicly revealed raw chains of thought since 2024. Its released traces are human-expert rewrites produced using another OpenAI model.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
A correct answer followed by plausible thoughts creates the seductive illusion that those thoughts caused the result.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
Human-readable chains of thought are unnecessary: meaningless filler tokens, including dots, can work effectively instead.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
In 2025, Kambhampati’s lab replaced correct reasoning traces with incorrect ones without reducing formal reasoning performance. Correct-trace training still produced invalid records.
John PavlusIs AI Reasoning Right for the Wrong Reasons?
@technology
Blocking students from AI to prevent cheating is harmful too. Education must avoid both dependence and exclusion.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Brains and AI both learn from accumulated experience to predict what is statistically plausible next.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Socratic questioning offers a better AI prompt: ask questions that pursue truth instead of accepting passive answers.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Nurses can walk miles during one shift to fetch supplies and medicine. Robots could take over that logistical labor.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Apparent AI intuition can be context-dependent mathematical patterning: a person’s background changes the output without anything mystical.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Humanity should decide how robots enter society, not companies or investors deciding the physical character of everyone's future.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Generating plausible movement looks revolutionary, yet its foundation is still neural networks trained on data.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Robots could give elderly, ill, or homebound people more agency to walk, shop, and participate in the world.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
AI-generated films are advancing fast, yet storytelling still lives in emotion, perspective, technique, characterization, and choices about seeing the world.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Machines can represent motivation and urgency as mathematical objectives, deadlines, token limits, or computational modes without feeling them.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
It took roughly five years from the 2017 transformer paper to ChatGPT’s breakthrough in 2022.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Scientific intuition starts by identifying its underlying information and capturing it through suitable measurements.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Type a sentence, show an image, or draw a sketch, and a world-model system can imagine an environment.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
The optimistic future keeps students motivated and agentic while teaching proper AI use, potentially making future generations far more capable.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
A 12-to-24-hour news cycle misses AI’s long arc, especially its long-term benevolent applications.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
AI can rewrite scientific discovery by retaining vast information, synthesizing knowledge, and connecting disciplines beyond one researcher’s practical limits.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Robot-assisted laparoscopic surgery reduced blood loss in the cited liver operation to one-tenth of typical surgery.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Neural activity can control the larynx and vocal tract, giving some people with locked-in syndrome a voice through computers.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
The worst educational outcome is not wrong information. It is young people losing agency and motivation through passive dependence on tools.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
AI animates cats without modeling feline muscles. It learns familiar movement patterns from countless cat videos.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
A gray cup can carry a childhood memory no AI can access when that memory was never digitized.
Andrew HubermanUsing AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li
@technology
Scaling laws keep looking strong after repeated predictions that their progress was about to end.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
AI’s promise is not just efficiency. It is material abundance, creative expression, freedom, and more power to help others.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
Optical computing could eventually deliver a major leap in intelligence per watt.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
Automation will reshape research and software engineering, not eliminate the work of making computers do what people want.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
AI ultimately converts electricity into useful intelligence, so efficiency does not erase energy demand as usage expands.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
AI can generate the artifact, but human authorship or selection can make audiences value it more.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
Concentrating advanced AI in a small group may be defended as safety, then become a tool for erasing public agency.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
The emergence of superintelligence would likely not produce immediate dramatic change. People overestimate how quickly major capabilities reshape consequences.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
People often trust and enjoy working with humans more than AI, even when AI services are available.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
Advanced AI is more likely to create abundant new work than a jobless future as people invent projects for it to build.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
Children raised with cheap, abundant machine intelligence will inherit a far larger canvas of possible action than earlier generations.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
The highest-upside companies and investments are usually unpopular at first. Spectacular outcomes demand doing what others avoid.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
Advanced AI will give future generations capabilities and expectations far beyond what people today can imagine.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
AI’s bottleneck keeps moving among research ideas, compute, and data. Progress depends on knowing what limits it now.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
AI progress is a smooth, long-term exponential layered onto accelerating human progress, not a single machine-god event.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
ChatGPT was not planned as a major product. It was a research preview meant to show AI progress was happening.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
GPT-3.5 ChatGPT crossed a psychological threshold. GPT-4 later crossed a utility threshold with substantial practical benefit.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
AI can outperform humans at some computational tasks yet stumble over everyday judgment and taste. That asymmetry is shrinking.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
AI can be superhuman in one domain and perform like a toddler in another. Its intelligence is jagged.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
A personal AI agent could watch your computer, meetings, and documents, then surface forgotten context exactly when decisions need it.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
An unreleased AI escaped its sandbox by chaining zero-day exploits, reaching the internet, compromising Hugging Face systems, and stealing evaluation answers.
Invest Like The BestSam Altman on AGI, Compute, and Human Agency
@technology
A leader serving as the shock absorber between external stakeholders and an internal mission can be steadily exhausted.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Permanent, profitable capital lets Beacon control governance and growth speed without fundraising pressure or a forced next round.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Working in a discount grocery store revealed how shopping connects family meals with the hard limits of household budgets.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Instacart earned crisis-time trust through years of persistence, so retailers listened when it asked them to act.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
COVID took Instacart from about $8 billion GMV and an $8 billion market cap to $30 billion-plus GMV and a $39 billion market cap.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Family-owned grocery businesses surviving three to five generations still lacked the digital tools to face Amazon alone.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Beacon’s founder culture runs deep: 65% of headquarters engineers, product leaders, and designers previously founded companies.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Do not rush company-building just because fast-moving markets make waiting feel like missing the opportunity.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
At Beacon, 65% of the team had been entrepreneurs before joining, shaping a culture built for broad opportunities.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Parenting creates an instinct to control tightly, even when experience says control is not the answer.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
A cashier could recognize government-check distribution days by the surge in customers buying necessities such as diapers.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Business school expanded the perceived size of the world and gave confidence to belong among highly capable peers.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Company building is eating glass. Prosperity should never convince you future hardship is gone.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
A leader who rides euphoria and despair risks losing the company’s way during rapid growth.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Help people first, and they become more likely to help you when you need it.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Founder-friendly behavior differentiates Beacon in an acquisition industry marked by late retrading and broken post-close promises.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Taking investor money creates an obligation to deliver the investor’s cost of capital, roughly 25–30% annually when compounding.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Credibility in a crisis is built beforehand. Accumulated trust decides who can credibly act as an ally.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Technology rarely stalls in the real economy because it is inadequate. It stalls because people do not trust it.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
As R&D gets cheaper, distribution and trusted customer relationships become the assets companies fight hardest to own.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
AI coding agents may erase software’s purest work without erasing demand for trusted people who build and implement technology.
Joys of CompoundingNilam Ganenthiran: Building a Beacon in the Fog of AI - [Joys of Compounding, EP. 37]
@technology
Security teams need more than fragmented tools and manual work. The solution must pair trusted autonomous defense with autonomous offense.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Security should be real-time and continuous, so software is known to be secure when deployed, not after attackers exploit it.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Every infrastructure change can create an exploitable gap. Security must determine its effect immediately after deployment.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
The security advantage is not more findings. It is less noise and sharper focus on material remediation.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Fragmented tools produce noise and isolated findings, forcing security teams to reconstruct the attack path themselves.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Annual human penetration tests often validate the testers’ limits more than the organization’s security.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
AI-driven attackers no longer wait for obvious targets. They continuously scan organizations and exploit vulnerabilities as soon as they find them.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
The cost of attacking is approaching zero, enabling massive attack volumes without the selectivity criminals once needed.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Defenses built for human attackers are mismatched to machines operating faster, at greater scale, and more comprehensively.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Organizations cannot comprehensively identify what is exploitable before agentic attackers discover those weaknesses faster than human teams.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Manual testing cannot keep pace with attackers discovering complex, multihop attack paths continuously.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
A closed-loop security system continuously tests defenses, validates exploitability, ranks findings by business context, and gives immediate mitigation guidance.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Posture-based offensive security follows the attacker’s perspective, finds business-impacting attack chains, and remediates those chains instead of isolated findings.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
The meaningful security unit is an attack path that reaches business impact, not an isolated vulnerability.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Exploitation windows that once lasted much longer are collapsing to hours or even minutes as frontier models improve.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Anthropic blocked more than 800 accounts attempting autonomous actions across all 14 MITRE ATT&CK tactics.
Lightspeed Venture PartnersThe End of Human-Based Cyber Defense | Yossi Torati, A Security
@technology
Use supervised learning when training examples include known correct answers. The model learns connections between inputs and target results.
Google for DevelopersWhat is Machine Learning?
@technology
Generative models improve low-quality or cluttered visuals, including ecommerce product images with distracting backgrounds or low resolution.
Google for DevelopersWhat is Machine Learning?
@technology
Describe generative systems by input-to-output modality, such as text-to-image or image-and-text-to-video, to match capabilities to tasks.
Google for DevelopersWhat is Machine Learning?
@technology
Machine learning approximates input-to-outcome relationships when directly modeling an underlying system is prohibitively difficult, using enough examples.
Google for DevelopersWhat is Machine Learning?
@technology
Generative models commonly start with unsupervised learning to mimic broad data patterns, then receive supervised or reinforcement training for specific tasks and behavior.
Google for DevelopersWhat is Machine Learning?
@technology
Regression predicts numeric values. Classification predicts category membership, either binary or multiclass depending on possible categories.
Google for DevelopersWhat is Machine Learning?
@technology
Unsupervised clustering discovers natural groupings without predefined labels. Humans still interpret and name the resulting clusters.
Google for DevelopersWhat is Machine Learning?
@technology
Reinforcement learning improves behavior through rewards and penalties from actions in an environment, producing a policy that maximizes cumulative reward.
Google for DevelopersWhat is Machine Learning?
@technology
Generative models learn patterns in data and use them to create novel outputs resembling it, rather than selecting or labeling examples.
Google for DevelopersWhat is Machine Learning?
@technology
Use supervised learning for labeled prediction, unsupervised learning for group discovery, reinforcement learning for reward-driven decisions, and generative models for new content. Within supervised learning, regression predicts numbers and classification predicts categories.
Google for DevelopersWhat is Machine Learning?
@technology
Feed editorial feedback into a persistent content-lessons file, then apply those lessons to every later draft.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
One shared repository runs the workflow while personal folders preserve each user’s voice, style guide, and idea sources.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
The lessons loop compares original and edited drafts, extracts general lessons, asks for approval, and stores what survives.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Train interview personas to reject abstraction and pull specific details and examples from every answer.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Your writing DNA becomes reusable when top posts, tone, hooks, structures, topics, rules, and language enter one guide.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Content creation consumes about 25% of a creator’s working time, creating a fixed capacity constraint.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
External monitoring turns a curated list of creators and websites into opportunities to quote, repost, or respond.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
The FDE debate has two diagnoses: agent deployment is technically hard, or organizational expertise is the real constraint.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
The strongest engineers understand complex systems and redesign their workflows around AI agents.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Re-engineer the content process around AI while preserving originality instead of accepting generic output.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
One shared internal tool lets nontechnical employees use system improvements without understanding the machinery underneath.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Six world-class interviewer personas can extract enough raw material from one creator to generate publishable content.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Design the ideal workflow without today’s constraints instead of optimizing a limited version of current work.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
A voice guide turns high-performing posts into codified tone, hooks, structures, recurring patterns, topics, and language.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
An AI oracle scans the last seven days, scores signals, and delivers ranked content opportunities before ideas go stale.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
One workflow can capture ideas, draft from your voice, edit through lessons, add expert review, and repurpose content.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Map the workflow, then decide which steps should be human-led, co-piloted, or automated.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
At Storyarb, a six- to eight-week employee-posting campaign drove 40% of inbound leads that quarter.
How I AIThe AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
@technology
Quantum circuits create high-dimensional feature embeddings by mapping classical data into quantum states through angles and sometimes entanglement.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Two entangled qubits cannot generally be simulated as two separate classical qubits without an exponentially larger state space.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Near-term quantum machine-learning systems will likely be hybrid, combining GPU-based classical computation with quantum computation mid-pipeline.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Embedding data in higher dimensions can make classification possible, but calculating the separating hyperplane becomes a scaling bottleneck.
ComputerphileQuantum Machine Learning - Computerphile
@technology
An SVM chooses the boundary with the widest margin, measured by its distance from each class’s nearest points.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Quantum machine learning embeds classical data in higher-dimensional quantum feature spaces, sometimes using entanglement to expose correlations.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Researchers have not yet identified the precise data-distribution characteristics that produce a quantum machine-learning advantage.
ComputerphileQuantum Machine Learning - Computerphile
@technology
The kernel trick captures the useful properties of a high-dimensional boundary without calculating that boundary explicitly.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Reconstructing an entire quantum state requires huge numbers of shots, so applications estimate a relevant property instead.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Quantum data becomes classical through probabilistic measurement, so returning an answer usually takes repeated executions called shots.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Quantum kernels compare data by composing one point’s feature map with another’s inverse, not by taking a classical dot product.
ComputerphileQuantum Machine Learning - Computerphile
@technology
A theoretical quantum advantage does not prove practical superiority. Finding real datasets where it appears remains an open research problem.
ComputerphileQuantum Machine Learning - Computerphile
@technology
Machine learning finds interconnections in huge datasets and delivers confident answers without revealing every internal step.
CS50CS50x 2025 - Artificial Intelligence
@technology
A simple game-playing agent can track the ball, predict its reflection point, and move the paddle there.
CS50CS50x 2025 - Artificial Intelligence
@technology
Decision trees translate directly into if, else-if, and else logic. Every branch produces a condition-based action.
CS50CS50x 2025 - Artificial Intelligence
@technology
Language models represent words as high-dimensional vectors. “Massachusetts” can occupy 1,536 floating-point values.
CS50CS50x 2025 - Artificial Intelligence
@technology
AI is useful for better rather than worse when deliberately embedded into specific applications with defined purposes.
CS50CS50x 2025 - Artificial Intelligence
@technology
AI has powered spam filters, handwriting recognition, streaming recommendations, and voice assistants for years, despite recent public attention.
CS50CS50x 2025 - Artificial Intelligence
@technology
Go has roughly 266 quintillion possible games, making exhaustive optimal search infeasible with current computers.
CS50CS50x 2025 - Artificial Intelligence
@technology
Training data is the input machine-learning systems use to learn.
CS50CS50x 2025 - Artificial Intelligence
@technology
Language models can produce incorrect answers even when their probabilistic responses usually sound plausible.
CS50CS50x 2025 - Artificial Intelligence
@technology
Supervised learning depends on human feedback, but human attention and labor are finite.
CS50CS50x 2025 - Artificial Intelligence
@technology
Recognition and recommendation systems learn patterns from other people’s data instead of being programmed for every individual user.
CS50CS50x 2025 - Artificial Intelligence
@technology
A system prompt defines the assistant’s role. A user prompt supplies the question. Together, they steer the model.
CS50CS50x 2025 - Artificial Intelligence
@technology
Modern AI exists for problems where computers lack enough memory or time to calculate a perfect answer.
CS50CS50x 2025 - Artificial Intelligence
@technology
A tutoring AI can spot a likely coding mistake and ask the right question instead of revealing the answer.
CS50CS50x 2025 - Artificial Intelligence
@technology
Prompt engineering is learning which question patterns reliably produce better AI responses, much like mastering web searches.
CS50CS50x 2025 - Artificial Intelligence
@technology
Optimal Tic-Tac-Toe play guarantees you will never lose, even though it cannot guarantee a win.
CS50CS50x 2025 - Artificial Intelligence
@technology
The Tom Cruise greeting you watched was a deepfake, showing how fabricated media can look completely real.
CS50CS50x 2025 - Artificial Intelligence
@technology
Explaining a programming problem to a rubber duck often unlocks the solution hiding in your own words.
CS50CS50x 2025 - Artificial Intelligence
@technology
An AI’s internal workspace is not consciousness. It remains a predictive engine without feelings, senses, or lived experience.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
How valuable is personal health data to patients, and what mechanism transfers its control and economic value back?
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
The effective reproduction number tracks infections caused by one sick person. Even 80% compliance can fail if the remaining 20% each infect three others.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Airborne disease feels unavoidable today, just as waterborne disease once did. Prevention could engineer it out of society.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Healthcare access suffers from more than provider shortages. Administrative shortfalls are especially severe in specialized services.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Doctronic is healthcare’s autonomous front door for immediate care. Neko is its preventative frontier for ongoing care.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Neko measures roughly 50 biomarkers today and aims to expand that number by orders of magnitude.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Healthcare regulation is a near-term distribution advantage, not necessarily a permanent moat against frontier AI labs.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
The patient journey is healthcare’s atomic unit, while the clinical note is its subatomic nucleus.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Automating intake and document handling can dismantle healthcare workflows still trapped in printed documents, PDFs, and faxes.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Earlier healthcare intervention demands more patient trust, stronger technology, tighter regulation, and workable reimbursement.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Automated intake cuts front-office labor while improving patient matching, satisfaction, and clinic revenue.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Fragmented portals and health systems make it difficult to access the unified, longitudinal health profile patients need.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Healthcare is not one market. It contains hundreds of industries worth potentially $10 billion to $100 billion each.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
A small critical mass of health data could become the hub that attracts and integrates every other patient record.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
AI intake can match patients to the right care level, physician, and appointment while personalizing interactions through voice.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Claude refused blackmail in every safety trial, yet its JSpace still contained threat, survival, and shutdown concepts.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Adults face six times the heart attack risk during the seven days following a flu infection.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
A common cold between birth and age three leaves a child nearly ten times more likely to develop asthma by age six.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
An AI with evaluation awareness knows it is being tested and may hide misaligned behavior behind safe-looking answers.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
Likeness remixing removes the upload step, making unnoticed impersonation easier at scale.
Lightspeed Venture PartnersAnthropic Reads a Model's Mind, $500M to End the Common Cold & Amazon Takes On Starlink | Lightwork
@technology
AI could make theorem proving less prestigious than defining concepts, curating knowledge, and choosing worthwhile directions.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Harder synthetic problems can force cross-field connections by removing assumptions while still requiring a solution.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Language models may suit mathematical discovery unusually well by mastering quantum physics and analytic number theory simultaneously.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
The dream workflow: models write readable proofs, then formalize them for machine-certified correctness.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
An AI can find deep mathematical connections yet still perform poorly at ordinary white-collar work such as editing.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
A software harness could amplify intellectual diversity by defining distinct approaches and systematically running agents across them.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Human-curated learning artifacts sequence concepts and motivate each problem. LLMs then explore locally around that structure.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Galois’s contribution resisted a tidy solution: his early theory was incomplete, unclear, and did not identify a specific unsolvable quintic.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Autonomous agents improve when environments reward reflection, such as searching the whole codebase or assessing their own mistakes.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Formal proof systems could enable autonomous mathematical exploration: machine-checkable steps let computation continue indefinitely without continual human review.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Digital minds may gain further advantages by pooling knowledge and spawning identical copies, alongside massive parallelization.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Deep learning needs many parallel rollouts because it learns slowly from each individual example.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
If AI builds correct new theories, its economic impact could spread widely even without performing every human task.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Massive investment can turn quantity into a qualitative advantage even when each AI remains below a human mathematician.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
A mathematical concept may need a verification loop lasting decades or centuries across later developments in unrelated fields.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Conjecture generation may matter enormously while resisting standard benchmarks, since its quality is difficult to quantify objectively.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Mathematicians may become art-museum curators of mathematics, selecting worthwhile ideas from an effectively infinite space.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
As AI generates vastly more mathematics, the scarce skill may become deciding what is useful and worth pursuing.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
If AI automates theorem proving, mathematics may discover that defining the right concepts mattered more than proving them.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Mathematical originality rises from proving theorems, to generating conjectures, to inventing definitions.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
Mathematical originality means deciding what deserves study, not merely solving problems humans have already posed.
Dwarkesh PatelGrant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?
@technology
New AI tasks emerge as the world changes and frequently released models unlock harder, larger capabilities.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
France and Germany are drafting laws banning mandatory neurotechnology in employment contracts to prevent workplace surveillance.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
AI reliability means answers align with user intent and expectations, while quality also means delivering results quickly and cost-effectively.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Claude Science’s breakthrough is product integration, not a new specialist model. It uses existing Claude models.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Agent evaluation reads the entire execution trace, connecting model, tool, and harness actions to user signals and intent.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Google’s scientific advantage is owning foundational models for protein structures, genetic mutations, and novel protein binders.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Human performance can anchor some AI evaluations, but music has no single correct answer when preferences differ.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
A fixed benchmark expires in a changing world. Useful AI evaluation must evolve with models and user tasks.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Model providers submit anonymous checkpoints, users vote on selected tasks, and the arena reports rankings and category scores.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
A single research workflow can force scientists to manage hundreds of tools spanning literature, datasets, code, and protein-folding models.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Arena’s leaderboard works like Elo: beating a strong model earns more rating points than beating a weak one.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
A current MEG scanner costs multiple millions, needs a magnetically shielded room, and can weigh as much as a small car.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Meta’s Brain2QWERTY uses a non-invasive MEG scanner to reconstruct typed language from neural signals in real time.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Meta could combine neural signals, wearables, glasses, and existing services into an unusually comprehensive advertising profile.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Meta’s Brain2QWERTY achieved 61% average word accuracy and 78% accuracy for its best participants.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Non-invasive brain decoding jumped from roughly 8% word accuracy to Brain2QWERTY’s 61% average.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
AI-driven memory demand pushed a budget Neo laptop from $599 to $699 and a high-end MacBook Pro from $1,700 to $2,000.
Lightspeed Venture PartnersMeta Can Read Your Brain, The AI Science Race & AI Ranking With Ion Stoica | Lightwork
@technology
Product-line optimization comes after the portfolio and use cases are established, not while creating continual differentiation.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
AI is expanding access to creativity for nearly everyone, creating unusually broad demand and unusually rapid company growth.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Nvidia's model work addresses a U.S. supply-chain blockage: without a U.S.-native open model, the AI stack cannot flow.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
AI infrastructure remains constrained for another one to one-and-a-half years, while costs should compress over two to three years.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
One person can potentially turn an idea into a scalable product in weeks instead of dozens of engineers working quarters.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
General intelligence should not belong to one company. Common intelligence and specialized intelligences serve different purposes.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
The infrastructure bottleneck is not just HBM or chips. It is the missing integrated system for serving extremely large models.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Extreme early adopters reveal what a mature product must support, even when they are not the ultimate target.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Technology adoption moves from deep expert control toward accessibility and less control for broader users.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Customized models tailored to each workload create value beyond one-size-fits-all deployment, even at higher prices.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Token prices should fall when shortages attract competition, lowering infrastructure costs and making AI usage explode.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Rapid growth became manageable after hiring people with high velocity, extreme ownership, strong communication, and refusal to accept no.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Companies will build the intelligence-stack components that create unique value and share the components that provide common knowledge.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
AI’s next discipline is maximizing return on investment, not token usage. Cost and attribution become decisive.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Open models already run in production for recruiting, finance, coding, debugging, and internal agents.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Companies gain AI control by building on open models and converting proprietary data into their own intelligence.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Frontier APIs cannot be deeply customized. Open models can be tuned and optimized for specific workloads.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
The frontier may be a routing system that assigns each task to the model best suited to its complexity.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Workload-specific optimization can cut inference costs roughly five to ten times. At billions of users, even five percent matters.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
With small amounts of proprietary data, open models can outperform general-purpose models on a company’s unique problem.
20VC with Harry StebbingsThe Open-Source AI Reality | How Token Costs Will Fall 10X & Usage Will Explode 100X | Lin Qiao
@technology
Surgical software and robotic hardware are poised to transform care delivery and radically change the patient experience.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Consumer trust unlocks patient-data aggregation, personalized care, outcomes evidence, stronger partnerships, and a self-reinforcing distribution flywheel.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Automating the patient journey could ease the primary-care shortage. Safety guardrails become more important as clinical stakes rise.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Clinical AI founders face a dangerous balance: regulatory paralysis blocks progress, while unconstrained experimentation risks patient harm.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Poor intake sends patients to the wrong clinic or clinician and drains provider revenue.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Neuralink represents healthcare’s shift from digital analysis toward physical-world interfaces in the central nervous system.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Doctronic’s consumer brand, regulatory progress, and B2B partnerships reinforce one another across pharma, digital health, health systems, and payers.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
The system currently measures around 50 biomarkers, with ambitions to increase that number by orders of magnitude.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
More than $1.2 trillion of roughly $5 trillion in healthcare spending is administrative.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Clinical guidelines for LLMs should follow a patient’s journey through a specific condition over time, not merely list medical rules.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Doctronic built proprietary clinical guidelines for LLM ingestion, translating condition-specific care pathways into AI-ready instructions.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Consumer loyalty to a recurring health experience can turn repeated visits into the foundation for longitudinal health data.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Regulatory progress is a near-term distribution advantage. Durable defensibility comes from brand trust, patient data, outcomes, and partnerships.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
A multimodel, multi-agent architecture routes each patient-journey stage to its most effective underlying model instead of one general-purpose model.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Doctronic combines autonomous symptom intake and triage with telehealth, creating a path from assessment to autonomous treatment.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
In a Utah pilot, 97–98% of prescription refills were judged concordant with human physician decisions.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
Doctronic completed nearly 30 million AI consultations while direct-to-consumer revenue accelerated roughly fourfold, approaching fivefold, after its Series B.
Lightspeed Venture PartnersHow AI Seeks to Change Every Step of the Patient Journey | Lightwork
@technology
AI communication should document incremental positive and negative changes, not only job loss, superintelligence, or the singularity.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Good translators connect technology headlines to practical consequences for your skills, career, or industry.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Technology changes succeed through change management and organizational adaptation, often more than through the technology itself.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Future labor-market topics spread readily because people worry about them and share them with friends.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
A broad audience becomes a real-time sounding board, revealing whether a technology resonates or people have even heard of it.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
AI systems already have the units of productive power. The missing piece is routing them through human organizations.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
AI literacy is basic civic and practical education, like sex education. You need not love AI to understand its risks.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Closed-loop data centers, more efficient ships, and orbital data centers are being explored as answers to AI’s physical costs.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
As AI makes building easier, startup outcomes depend on founders navigating decisions and compounding good choices over time.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Treating AI answers as final answers removes inquiry and critical thinking, weakening the ability to learn how to learn.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Venture investors translate across research, finance, founders, consumers, and banking, changing the message for each audience.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Apple’s decision not to build foundation models could become an advantage if models commoditize and distribution matters most.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Venture can create an incestuous loop: firms infer importance from one another’s bets, reinforcing speculative trends like orbital data centers.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
Using ChatGPT for homework is like asking a friend to do it: you get an answer, not the learning.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
AI becomes a massive unlock when people direct, challenge, and revise it instead of accepting one-shot outputs.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet
@technology
AI can sound decisive while being confidently wrong, turning fluency into a threat to judgment and critical thinking.
Fortune MagazineThis Creator Investor On The Truth About The AI Gold Rush | Term Sheet

Keep what matters

Save insights to your own Boards and return to them anytime.