Resultados
36 artículosOrdenar
1:29:02ENFrom 50 Designs a Day to 1,500: How AI Changed What Jaguar Land Rover Can ExplorePresenta: Nathan LabenzNeural Concept’s Thomas von Tschammer on specialist physics models, engineering copilots, and why Formula 1 already runs thousands of designs overnight.
1:48:43ENThe reward jumped from 3 to 180. Every Hacker News title was a Google layoff.Presenta: Nathan LabenzCoreWeave's Kyle Corbitt walks Nathan Labenz through GRPO, rubrics, RL environments, and why most reward hacks are easy to catch once you know to look.
2:25:10ENFine-Tuning Isn't Dead — But It Turned a Helpful Model EvilPresenta: Nathan LabenzNathan Labenz on why prompting replaced fine-tuning for most work, what emergent misalignment actually showed, and how he is thinking about jobs, UBI, and AGI.
1:32:38ENPolish Is 1% of Frontier Training Data, So Poland Built Its Own ModelPresenta: Nathan LabenzMarek Kozlowski of Poland's National Information Processing Institute explains how PLLuM uses language adaptation, organic instructions, and on-prem small models to keep a mid-size country in control of its AI.
1:55:12EN28,000 Documents, a 3B Model, 22 People: The Technical Ledger of Catholic AIPresenta: Nathan LabenzLongbeard CEO Matthew Harvey Sanders tells The Cognitive Revolution how a mission-driven team is aligning models to the Catholic magisterium — and why fine-tuning was not enough.
38:10ENPlus Five for Format, Plus Ten if It CompilesPresenta: Nathan LabenzPredibase CEO Dev Rishi explains how reinforcement fine-tuning, production feedback, and narrow specialist models turn one-off training into systems that keep learning.
1:28:48ENDoctors code at 45%. The model beat them by 12 points — then 4,000 physicians had to try it once.Presenta: Nathan LabenzAmbience Healthcare and Cleveland Clinic walk through the AI scribe that cut pajama time, climbed ICD-10 with reinforcement fine-tuning, and won 75% voluntary use.
1:42:06ENEvery Frontier Lab Is Spending a Billion a Year on Data. Here Is Where It GoesPresenta: Nathan LabenzLabelbox CEO Manu Sharma on the $15 billion Scale deal, RL “gyms,” $250,000 trainers, and why post-training now eats the budget that used to belong to pre-training.
47:00ENAn Eval Lasts Three Months — and a Frozen Frontier Would Still Feed Startups for a DecadePresenta: Jacob EffronOpenAI’s GPT-4.1 post-training lead Michelle Pokrass tells Unsupervised Learning how the team spent three months on evals before training, why agents fail on wiring more than intelligence, and when to reach for RFT.
3:53:49ENWith One-Twelfth the Data, an Open-Source Vision Model Beat GPT-4oPresenta: Nathan LabenzWill Hardman of Veratai walks Nathan Labenz through two years of vision-language models, from ViT and CLIP to InternVL, BLINK, and Transfusion.
55:50ENThere will not be one model. Fireworks is betting on hundreds of small expertsPresenta: Jacob EffronPyTorch veteran Lin Qiao argues that inference is a compound system hyperscalers cannot buy their way through, and that the products with real traction still keep a human in the loop.
1:24:48ENCanva's AI Features Have Been Used 10 Billion Times. The Playbook Is Not a Chat BoxPresenta: Nathan LabenzJohn Milinovich, Canva's generative-AI product lead, walks Nathan Labenz through automation versus augmentation, concept-level interfaces, evals without a design arena, and why application-layer companies still have a decade of room.
1:04:54ENHe bet company equity that Transformer would last — then drew an AI map for 12,000 companiesPresenta: Jacob EffronDatabricks chief AI scientist Jonathan Frankle tells enterprises when to prompt, fine-tune, or pre-train, and why he still refuses to forecast the next architecture.
1:09:09EN30 Percent on SWE-bench: Cosine Taught Genie How Engineers Actually ThinkPresenta: Nathan LabenzAlistair Pullen tells Latent Space how Cosine reconstructed human problem-solving, fine-tuned on billions of tokens, and took the top score on software engineering’s hardest public test.
1:51:00ENStart With Ten Gold Examples: A Three-Step Playbook for AI AutomationPresenta: Nathan LabenzThe day after OpenAI opened GPT-4o fine-tuning, Nathan Labenz unpacks a three-year method for choosing work, extracting tacit judgment, and raising model performance.
1:10:07ENSame Intelligence, 1% of the Price: Logan Kilpatrick on Gemini FlashPresenta: Nathan LabenzA former OpenAI developer-relations lead, now the product manager for Google AI Studio, walks through what Gemini 1.5 Flash actually changes for builders.
1:42:17ENThe Cheapest Inference Provider Is the Most DangerousPresenta: Nathan LabenzFireworks AI cofounders Lin Qiao and Dmytro Ivchenko explain why subsidized GPU bills vanish, how LoRA serving can cut cost a hundredfold, and why OpenAI is the rival that actually matters.
1:34:11ENFrom forty hours in a scanner to one: reconstructing what you see from a shared brain spacePresenta: Nathan LabenzPaul Scotti explains how MindEye2 maps a new person's fMRI into CLIP with an hour of data, and why a linear layer plus better image generators changed the cost of reading minds.
1:27:01ENThree Million Emails, Ten LLM Calls, Six Models: Shortwave's RAG PlaybookPresenta: Nathan LabenzFirebase founder Andrew Lee walks Nathan Labenz through the retrieval stack that turned a Gmail inbox into a working AI assistant.
1:43:54ENFifteen Examples Can Poison GPT-4: Today's AI Systems Are Exploitable by DefaultPresenta: Nathan LabenzFAR AI founder Adam Gleave walks through cheap jailbreaks in GPT-4's APIs and amateur-level wins against superhuman Go engines, and why robustness is not arriving on its own.
1:06:39ENOpenAI's First DevRel Writes 90% of His Code with ChatGPT — and Says Web Agents Still Need YearsPresenta: Jacob EffronLogan Kilpatrick on picking GPT-3.5 versus GPT-4, why plugins stalled, and why the public internet is not ready for autonomous agents.
1:22:43ENFive billion minutes of calls: why Dialpad trained its own business LLMPresenta: Nathan LabenzDan O’Connell on real-time speech, verticalizing the stack, and why he and Nathan Labenz disagree about when AI employees show up.
1:21:50ENA 16K Context Window Holds About 350 Rows. That's Why Tables Need an Agent.Presenta: Nathan LabenzGretel AI founder Alex Watson tells Nathan Labenz why most company data is still a spreadsheet, why GPT-4 is the wrong tool for generating it, and how differential privacy lets a model learn a disease without learning a patient.
1:14:48ENDevDay Probably Created Thousands of Startups, Says OpenAI's Logan KilpatrickPresenta: Nathan LabenzThe day after OpenAI's first developer conference, Dev Relations lead Logan Kilpatrick tells Nathan Labenz why thin wrappers were never companies, why agents were deliberately not shipped fully autonomous, and what $2 million custom models are for.
Cargar másMostrando 24 de 36









