Loading activity…
@hamelhusain
Evals Evals Evals @OpenAI About Me: https://hamel.dev
AI
SWE-bench Multimodal v2's issue text and code usually contain enough information to solve tasks, the researcher says.
AI
Prime Intellect says revisions to AutomationBench’s tool descriptions, tasks and graders drastically changed model scores.
AI
Vik Paruchuri argues that scoring flaws can penalize correct OCR results and hide important differences between models.
YouTubeAI
Subvocal claims non-contact sensing decodes nearly invisible mouthing into text at more than 200 words per minute.
XAI
The authors recommend putting distinguishing words first and keeping structure and vocabulary consistent across items.
AI
Shreya says readers can order it on Amazon or get a copy at a book signing.
AI
Epoch AI says four benchmarks were verified and two lacked enough information for a review.
AI
Interconnects Podcast says the discussion asks why cyber and AI experts talk about recent events so differently.
AI
LangChain says its three agents handle sales leads, code reviews and marketing drafts, with many runs triggered in the background.
AI
In notes on their discussion, Harry Stebbings says Naveen Rao predicts a $10 trillion inference market within 10 years.
AI
A co-creator set October 30, 2026, as the deadline for product managers, engineers and designers to submit workflows.
AI
A post says it has 40-plus submissions and saturates only if a person heavily tunes agents' prompts to overfit tasks.
AI
The poster is looking for discoveries for an annual gift guide, but rules out Claude and ChatGPT.
AI
A user said they’d spent two days struggling with Codex’s auto-reload and asked what was worth trying instead.
AI
Raindrop says the feature watches every pull request and sends Slack DMs about how agent changes performed in production.
XAI
An event listing says authors Hamel Husain and Shreya Shankar will discuss designing AI evaluations and interpreting results.
AI
A user’s timeline moves from data analyst in 2003 to data scientist in 2010 and machine learning engineer in 2016.
AI
Sankalp Shubham found little change in resolved-task rates, while both models faltered on a separate visual test.
AI
The team behind Holon-S1 says it beat two other models at predicting spatial gene activity in a controlled comparison.
XAI
Hamel Husain and Antaripa Saha compare models on 100 leasing-agent traces, warning against an unchecked 0.50 threshold.
