Almost every ML personal gets confused you tell them that maximizing a convex function is NP-Hard for this reason
RESEARCH
-
Soumith discusses AI trends and thoughts on podcast
By
–
I did a fun podcast with @swyx and @FanaHOVA .
Listen (or read the auto-generated transcript) here: https://
latent.space/p/soumith I gave raw unprepared thoughts on a bunch of topics, hope you find them interesting enough to critique but not bad enough to dunk (ymmv). -
Analyzing AI Model Capabilities Through Training Data Examination
By
–
As always, when you try to analyze the capabilities of these models, please look at the datasets they’ve been trained on! https://
x.com/hahahahohohe/s
/hahahahohohe/status/1765435151817830834
… -
AGI Definition Problem: I Know It When I See It
By
–
So AGI is like pornography: https://
corporate.findlaw.com/litigation-dis
putes/movie-day-at-the-supreme-court-or-i-know-it-when-i-see-it-a.html
… -
Vec2text: Inverting Text Embeddings Through Stories
By
–
wrote a little blog post about our vec2text, our method for text embedding inversion! it's like the paper but told as a story: with a lot less jargon, and no math. check it out 🙂
-
Claude 3 Demonstrates Higher IQ Than Competing AI Models
By
–
I would also say (and forgive me for not having the data here) that Claude 3 has a higher IQ too, from what I have seen than any other models. I would say probably around 100.
-
Long Context and Memory Enhance AI Understanding
By
–
Long context, together with almost perfect retrieval and memory, would do that, plus sprinkle in a good level of understanding.
-
Microsoft CTO Discusses Generative AI Architecture at Stanford
By
–
Packed house at today’s @StanfordHAI seminar with @Azure CTO @MarkRussinovich. We’re diving deep into Microsoft’s AI architecture, as well as advancements and opportunities in trending AI research. You can still join via Zoom: https://
hai.stanford.edu/events/mark-ru
ssinovich-generative-ai-cloud-inside-microsoft-ai-innovation
… -
Improving LLM Testing Methodology Through Feedback
By
–
Appreciate the feedback. I'm taking notes so I can make future LLM tests even more thorough. 🙂
-
Silent Bugs in Fine-tuning: Attention to Detail Matters
By
–
Beautiful work / attention to detail trying to get Gemma to finetune correctly. There are so many foot guns here to be super careful with. All of these issues don't throw any errors, they silently make your network worse. A great example of what I wrote about in my "A Recipe for