amazing what we can learn when we share evals in public
@swyx
-
Asking Opus to Double-Check Its Own Answers
By
–
have you tried simply asking opus to doublecheck its own answers
-
Managing Secondary Metrics Beyond Primary Success Indicators
By
–
i dont know this system as well as you but its usually “take the most common failure mode that the primary metric doesnt cover” and make sure that doesnt increase”. in your case the problem is so well scoped that that may not actually exist/matter. in which case the subquestion
-
Paired Metrics Philosophy: Balancing Growth with Quality
By
–
i used to have this mgmt philosophy – “all metrics should come in pairs”. one to number go up, other to keep you honest in this case it feels like there should be a counterpart to ensure the experience/quality of inbound doesn’t deteriorate
-
Distinguishing Genuine AI Conference Organizers from Buzzword Chasers
By
–
looking at {competing ai conferences} its just stupidly obvious whether or not the organizers running the thing know whats up vs just chase buzzwords. trying to thread multidimensinal needles in a haystack here.
-
AI Engineer Conference Evals Track Receives Overwhelming Applications
By
–
going by applications the Evals track at this year's @aiDotEngineer conf is going to be INSANE. i have enough to just run a standalone EvalsCon lmao
-
Utility Belt: A Wearable for Wearables with AI Companions
By
–
Introducing Utility Belt — a wearable that wears other wearables. / test nRF52840 3dprint wrappers
Juggle up to AI gfs with 5min sunlight viewing
Who watches the watchers? Other watchers ofc To learn more, join launch party: https://
lu.ma/personal-ai (is real -
Vercel Uses AI Honeypots to Fight Bots and Reduce Costs
By
–
In response to bot traffic and scrapers, we (@vercel) have been using our AI apps, examples, and starter kits as security honeypots to develop new anti-bot + fingerprinting + rate limiting techniques. In tandem with our WAF product, we've since cut our daily OpenAI bill by 100x x.com/elonmusk/statu…
-
Family Tree of Transformer Models Evolution and Convergence
By
–
A beautiful "family tree" of large Transformer models from Vaswani et al's 2017 design So cool to see the lineages from GPT, BERT, T5, and PaLM develop and Diffusion models starting from a different place but converging with CLIP and ViT.
-
AI Notes: Image Generation and Reverse Prompt Engineering Resources
By
–
see if these help https://
github.com/sw-yx/ai-notes
/blob/main/IMAGE_GEN.md#img2prompt—reverse-prompt-engineering
…