@karpathy The concept of apps to me is basically a nice UI with backend to manage data and services.. if LLMs can just generate images and video on the fly and have services to interact with databases (coz that’s needed to store states), these agents can replace all apps.
Helpful update for students, you can now take full practice SATs for free in the @GeminiApp.
It uses vetted content from @ThePrincetonRev and gives you feedback straight away. Starting with the SAT today, but more tests are on the way!
For AI to be truly useful, it needs to understand you.
With Personal Intelligence, we’re beginning to solve this. With your permission, Gemini can now securely reason across your own data to answer questions that generic models simply can't - like suggesting plans based on travel dates in Gmail or your hobbies found in Photos. An exciting step towards a digital assistant that’s uniquely helpful to you.
🚨The Gemini 3 launched leaked hours before the official launch!! The model is INSANELY good, but let's dig deep and see what it means for the AI future.
1. It's not going to be super expensive!
Google trained this from scratch on TPUs as an MoE with 1M input and 64k token output. MoE means they can afford to make this cheap.
2. This is next level for Computer Use and automating real knowledge work.
ScreenSpot Pro is a very low-key benchmark which tests understanding screenshots in apps including AutoCAD, PhotoShop and many more. Gemini 3 is 2x SOTA on this with 73%.
3. Best AI for Math by far.
Gemini 3 is RLd on a ton of theorem proving data and is flawless on AIME and 23% on MathArena. Other models barely scrape 1%. There are clearly some generalized reasoning benefits given the sota 30% on ARC AGI 2 as well!
4. Insanely good at coding but unclear if best.
Live Code Bench elo of 2400+ is fantastic yet the one benchmark it doesn't get SOTA at SWE Bench. It's #1 on tool calling and terminal bench too.
Google pulled out everything they had on this one. They perfected their training recipe, trained on all the proprietary data they had and crushed every benchmark they could. This is a meaningful step up in model quality and clear proof that AI progress isn't slowing down.
I just don't know how any other company but Google wins on model quality here: they have a) cost advantage due to owning hardware b) ooms more proprietary data c) more disposal cash to spend on training and new data d) as good talent as anyone else. Flaunting their total dominance.
I suspect it will be hard to beat these results in the next 6 months.
I am watching a movie clips about @elonmusk's Tesla Semi - remember, the truck that was to revolutionize the transport industry when it entered the market 4 years ago 🤡
And I am going to tell you'all why it is a completely stupid vehicle. And I won't even want to talk about (1)
When confronted, Mahir threatened to put my luggage on hold if I do not accept his rudeness and just board the flight. I strongly suggest you take a survey of onboarding process with all other passengers, there are many such examples of rude behavior at the gate. Disappointed ☹️
@qatarairways : please train your on boarding crew at DFW, worst experience ever traveling with infant. Flight QR730 gate D15 on 24th Nov. Your crew namely Mahir Chowdhary and a lady asked me to QUIET my infant when he was crying and I wanted some assistance with onboarding… 1/2
The #NobelPrize committee couldn't reach Paul Milgrom to share the news that he won, so his fellow winner and neighbor Robert Wilson knocked on his door in the middle of the night.
#GoogleWorkspace embodies our vision for a future where work is more flexible, time is more precious, and enabling stronger human connections becomes even more important. https://t.co/WIA1vogQ4c
Was on @AC360 w/ @andersoncooper & @drsanjaygupta last night
Part 1 — higher education has raised prices 3x higher than inflation (150% vs. 54%). Higher ed is in for a massive disruption.
I think all human rights warriors, bollywood attention seekers and pseudo liberal journalists have gone to the UN to seek a referendum on whether policemen have human rights or not. Sick!
#Gujrat#Lucknow#Delhi@MamataOfficial@ArvindKejriwal
@MichaelKors : check out @gucci handing out water and snacks to people standing in Gucci line.. right in front of all the crowd standing in MK line. This is real customer service by @gucci
@united : how do I go about rescheduling my flight? No one on your customer service is picking up. You guys cancelled my flight last minute and I am stuck here 😞