Machine learning is the art of doing the wrong things that correlate with the right ones.
RESEARCH
-
OpenAI researcher departure highlights ongoing AI safety challenges
By
–
i'm super appreciative of @janleike
's contributions to openai's alignment research and safety culture, and very sad to see him leave. he's right we have a lot more to do; we are committed to doing it. i'll have a longer post in the next couple of days. -
What Makes Us Unique in the Age of AI – Starmus Presentation
By
–
That was a big stage … Notes on my #Starmus presentation “What makes us unique in the age of AI” available here: http://
inverseprobability.com/talks/notes/wh
at-makes-us-unique-in-the-age-of-ai.html
… -
Gemini 1.5 Pro Achieves Record 91.1% on MATH Benchmark
By
–
One other thing in the updated Gemini 1.5 Pro report: we show how a research model that is a mathematics-specialized version of 1.5 Pro achieves a record score of 91.1% on the MATH benchmark (the SOTA just 3 years ago, in May, 2021 was 6.9%!).
-
Brain Computing Power Compared to NVIDIA A100 GPU
By
–
The naive napkin math would go something like 1 brain ~= 1e11 neurons * 1e4 synapses * 1e1 fires/s = 1e16 FLOPS (i.e. 10 petaflops) NVIDIA A100 = 312e12 peak FLOPS, in-practice achievable utilization may be let’s say 50%, i.e. 156e12. Dividing you get 1 brain ~= 1e16 / 156e12 =
-
Responsible Path to AGI: Beyond Black and White Thinking
By
–
Jan, Ilya, etc. are not doomers. Take a look at their previous work. They believe in the benefits of getting to AGI, but don’t want to be reckless in how we get there. Not everything is black and white.
-
Extended Machine Translation Evaluation Results in Updated Report
By
–
The updated report is now 153 pages, and has quite a few new results. In the February report, I found the results on Kalamang translation for the Machine Translation from One Book benchmark quite exciting. In this updated report, we’ve extended this line of evaluation to test
-
Google Releases Gemini 1.5 Pro and Flash Models
By
–
Gemini 1.5 Model Family: Technical Report updates now published In the report we present the latest models of the Gemini family – Gemini 1.5 Pro and Gemini 1.5 Flash, two highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information
-
Medical Professionals Experiment with AI Before Consensus Established
By
–
Reality starts to set in: "SAN FRANCISCO — Surging interest in generative AI among medical professionals since ChatGPT’s launch is perhaps a testament to its potential — but it could also lead clinicians, and patients to experiment with it before there is wider consensus on how
-
OpenAI Dissolves Superalignment Team Tackling AI Dangers
By
–
My latest for @WIRED
: how OpenAI’s superalignment team, designed to tackle the dangers of supersmart AI, has been disbanded: