sigmoid.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A social space for people researching, working with, or just interested in AI!

Server stats:

595
active users

#DeepMind

4 posts4 participants0 posts today

"In the hopes that these systems are someday released to outside review, here are some questions that we hope some day will be answered.

- How do these systems work and how were they trained? What allows them to do so much better than earlier systems? Do they come with a cost in solving other sorts of problems? In the case of OpenAI-IMO, do the "new techniques" employed at all explain the stylistic weirdness?

- What is the scope of these systems? Do they generalize to other kinds of mathematical problems, to scientific problems, to more general classes of problems? Or can they be extended to deal with these? Or would they be a useful component in a larger system that dealt with these?

- What was the cost per problem, both for inference time (clearly high) and any domain-specific expertise required for training/augmentation etc and what would the economics of using these models be when it is released?

- Are these system compatible with tools such as computational tools, coding, and web search?

- What did the two systems do with problem 6, the one they couldn’t correctly answer? Did they give up, did they produce a partial answer that was in the right direction, or did they hallucinate a nonsensical answer?

Overall, the new work certainly might be exciting. Until outside scientists get to dig for a serious review we won’t really know for sure what it all means."

garymarcus.substack.com/p/deep

Marcus on AI · DeepMind and OpenAI achieve IMO Gold. What does it all mean?By Ernest Davis

🧬 What if your algorithm could breed itself into something smarter?
AlphaEvolve just cracked a 56-year-old math record—and it wasn’t coded by humans.
Google’s new AI doesn't just solve problems. It invents new ways to think.
Read how it saved $1B, outperformed 75% of human math benchmarks, and evolved a new logic.
👇 Dive in:
medium.com/@rogt.x1997/the-alg

#AlphaEvolve #AIrevolution #DeepMind
medium.com/@rogt.x1997/the-alg

Medium · The Algorithm That Bred Itself: AlphaEvolve’s Frightening BrillianceBy R. Thompson (PhD)

Imagine being a brilliant physicist/mathematician and still avoiding the most important problems because your career depends on publishing frequent papers, not solving the biggest mysteries in the world.

That's why you can't do things like this in academia.

english.elpais.com/science-tec

🧠 #DeepMind ha presentato #Gemini Robotics On-Device, un modello #AI avanzato capace di operare localmente su robot, senza connessione internet. 

👉 I dettagli: linkedin.com/posts/alessiopoma

___ 
✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: bit.ly/newsletter-alessiopomar 

Ny Teknik förklarar: Google knäckte arvsmassans 'mörka materia' – med hjälp av ai.

Google Deepminds Alphagenome använder AI för att förutspå hur DNA-mutationer påverkar genaktivitet. Genombrott för sjukdomsforskning och läkemedelsutveckling.#generativai #dna #deepmind #biologi #artificiellintelligens #ai #topp #tech #google
Alphagenome: Googles AI revolutionerar genetisk forskning

nyteknik · Alphagenome: Googles AI revolutionerar genetisk forskningBy Peter Ottsjö

Google bringt mit Imagen 4 Ultra sein bisher stärkstes KI-Bildmodell – entwickelt von DeepMind, abrufbar über die Gemini API. Vor allem bei Text in Bildern soll es neue Standards setzen. Ist das der nächste Schritt Richtung generative Perfektion – oder nur ein Hype mehr? #GoogleAI #DeepMind #KI 👇
all-ai.de/news/news24/imagen-4

All-AI.deImagen 4 Ultra: Googles KI-Bildgenerator wird jetzt ernstVon Felltexturen bis Typografie – Imagen 4 sorgt für Aufsehen. Doch ist sie besser als OpenAI und Midjourney?