Browse entries 1609–1632 of 1636. Return to the first page to search and filter the complete collection.
Play video
Quantize a single number in a model and it gets 20% dumber. That finding, from the super weights paper, is why Daniel Han's claim is less absurd than it sounds: GLM 5.2 goes from 1.5 terabytes to 250 GB, 86% smaller, without being 86% dumber. Layers are wildly unequal.
Play video
In 1945 Vannevar Bush described document scanning, OCR, speech to text, hypertext, search engines, a head mounted camera, and voice interfaces, in a single essay, before any of it existed.
Play video
Run terminal bench on Opus and on Haiku and Opus scores about three times better at a tenth of the cost, even though Haiku is far cheaper per token.
Play video
*Note: Kenton has just released Cloudflare OS today: https://x.com/KentonVarda/status/2084990137180590572 This talk was recorded a month prior to launch.* Claude needed a strikethrough the slide app did not have, so it added one to the app.
Play video
Every time a model launches there is a gap between the benchmark numbers and what the thing can actually do, and Nick Heiner argues the existence of the word benchmaxxing is the tell.
Play video
Ross Taylor opens with some history: back in 2022 he worked on Galactica, an early large model for science that briefly crossed the Rubicon on curated high quality data and intermediate reasoning tokens before the reaction overshadowed the work.
Play video
Swap compute for data on the scaling curve and the same money buys a better model, which is why Ari Morcos calls data quality the compute multiplier and the most underinvested part of training.
Play video
The next step after a model ships is teaching it to keep learning on the job, and Raymond Feng lays out how Applied Compute trains custom models with reinforcement learning that plug into whatever harness an enterprise already runs.
Play video
Ali Khial took three of the best engineers at G2i, pointed them at popular coding benchmarks, and hit a wall of tasks that were either too ambiguous to grade or quietly broken.
Play video
When a company is about to spend $100 million, Shawn Chan is the person in the room deciding whether it happens, and after fifteen years and roughly 200 investment committees he can spot the flaw fast. Now some of the decks in front of him are AI generated: cleaner formatting, never defensive, and confidently wrong.
Play video
Ask any LLM a financial question about your business. You'll get a fluent, confident, generic answer — one that doesn't truly know your business, or what happened when businesses like yours made that same decision. We build financial AI at Intuit serving 100M+ customers.
Play video
In finance a number is worthless until you can say where it came from. Vinoo Ganesh, CEO of Kepler, starts from the fact that language models are probability machines, brilliant at next token prediction and unreliable at the deterministic work, like arithmetic, that finance actually runs on.
Play video
At Palantir, forward deployed started as a literal description: you were deployed, physically, at the customer's site, and the onboarding project was keeping the platform from falling over. Natalie Meurer's dirty secret is that the title never settled after that.
Play video
Look at public software companies and very few ever crack half a million dollars in average contract value, and that gap is where forward deployed engineering lives.
Play video
Type llama into a catalog of 3 million public models and the result still has to feel instant. At 20,000 models any query is fast; at Hugging Face's scale, 14 million users and a million datasets on top, search becomes the hard part.
Play video
Data Quality Research at Prime Intellect and State of Data Author. Prior investor at Hummingbird and Costanoa.
Play video
A generated clip where the character stands frozen for four seconds can still score well, because the judge rewarded the gloss and the vibe instead of what actually happened.
Play video
Every month it is the same trap. A reasoning model gets upgraded at the same price per token, then quietly burns three times the output tokens. Or the new version costs 40% more and deprecates its predecessor in four months. Are you growing 40%? Making three times the revenue?
Play video
By declaring a task's inputs and outputs without initially considering model capability, you create the space needed to determine execution later. DSPy's promise is that AI engineering should happen above a particular prompt template or provider API shape: the Signature.
Play video
Live from San Francisco, AI Engineer World’s Fair 2026 continues with Day 2 of session programming from the main stage. Watch live for keynote sessions, main-stage programming, and more from World’s Fair 2026 as AI Engineer brings another full day of AI engineering content to viewers online.
Play video
Live from San Francisco, AI Engineer World’s Fair 2026 wraps with the final day of main-stage programming. Watch live for keynote sessions, featured talks, and closing-day highlights from World’s Fair 2026 as AI Engineer streams the final day of the event online.
METR’s November 2024 RE-Bench release provides seven research-engineering environments, human-expert attempts and agent transcripts. Tasks involve objectives such as optimizing kernels, recovering model performance and fitting scaling laws under specified resources. Results depend strongly on time allocation: agents often benefit from selecting the best of many short attempts, while humans benefit from longer continuous work.
Play video
April 20, 2026 - all times in EST -- 9:00am - Welcome to AI Engineer Miami -- 9:10am - Gabe Greenberg, G2i Opening Remarks -- 9:15am - Dax Raad, OpenCode Keynote -- 9:40am - Dexter Horthy.
Play video
AI Engineer session on Buy Now, Maybe Pay Later: Dealing with Prompt-Tax While Staying at the Frontier - Andrew Thomspson. It adds practical context for how teams are building and operating AI systems in production.