- The Neuron
- Posts
- šø AI aces Math Olympics
šø AI aces Math Olympics
PLUS: Apple's AI debut is getting delayed...
Welcome, humans.
Anyone watch the Olympics this weekend? We did, and we couldnāt help but notice quite a few AI ads, including this one from Google and multiple ads from Salesforce. It was like an AI extravaganza. If this reminds you of that one year crypto took over the Superbowl, thatās kinda how it feels to us, too.
Yāknow those charts that show where weāre at in a technology hype cycle? You can throw all those out. All we need is an analysis of x number of āinsert technology hereā adverts per hour for whatever the yearās main major sporting event is, and youāll know exactly where itās at in the cycle.
Hereās what you need to know about AI today:
DeepMind's AI secured silver at the International Math Olympiad.
Video game performers announced a strike over AI concerns.
Morgan Stanley launched new AI tools for client meetings and research analysis.
Apple delayed its āApple Intelligenceā AI rollout to October.
DeepMindās new AI just won a silver medal at the (math) Olympics.
While the world turned its eyes to the Paris Olympics this weekend, another medal-giving competition quietly made history. Two of Google DeepMind's new AI systems, AlphaProof and AlphaGeometry 2, achieved a silver medal performance at the International Mathematical Olympiad (IMO).
Hereās how the IMO works:
The IMO challenges pre-college mathematicians with six brain-busting problems over two 4.5-hour sessions. Think the hardest AP Calculus problem, but 10x harder.
DeepMind's AI nailed 4 out of 6 problems, scoring 28 out of 42 pointsājust one point shy of gold.
Notably, the AI solved the most difficult problem, which only five human contestants managed to crack.
Context is key: While the human contestants had ~9 hours, the AI solved one problem in under 19 secondsābut took up to 3 days to solve the others.
OK nerds, whatās the big deal?!
Well, 18 months ago, experts thought this level of math skills from an AI was 5-7 years away.
DeepMindās AI can now solve 83% of historic IMO geometry problems from the past 25 years, up from 53% in its previous version.
Many of these problems require creativity, reasoning, and problem-solvingāabilities once considered uniquely human. One judge (a medalist himself) was so impressed by the systemās solution, that he called it a āmagic key.ā
Some AI researchers see a gold win at the 2025 IMO competition as evidence that super-powerful AI is coming sooner than we thought. And after the silver medal news broke, predictions for this outcome surged from 25% to a 70%.
And if that happens, it could mean the main hurdle to transformative AI isn't new fundamental algorithmic breakthroughs, but good ol' fashioned engineering work.
What's next: DeepMind plans to bring AlphaProof/Geometry 2ās math wizardry to their Gemini models āvery soon.ā This means AI chatbots that are really good at math, an area in which theyāve previously struggled (it is called a ālanguageā model, after allā¦).
FROM OUR PARTNERS
Top sales teams still crush it in this economy. Attention is their AI secret.
Most sales reps are fed up with this economy (and arenāt you?). Too many deals are lost due to long sales cycles and budget cuts. Attentionās AI assistants can help.
Over 100+ sales teams from major companies likeā¦
Snowflake
Datadog
Stripe
ā¦all use Attentionās new AI Assistants to help close more deals and generate more revenue.
Check out one of our faves: Dax, a data analyst AI assistant, who:
Transforms call data into actionable insights
Identifies successful strategies for conversion
Creates performance visualizations for easy reporting.
If your business sells anything, we highly recommend checking Attention out here.
Around the Horn.
Video game performers announced a strike over concerns that they wonāt be protected against AI from major game studios.
Morgan Stanley launched a second AI, Debrief, which transcribes and summarizes client meetingsāand its āLLM Suiteā for research analysts is now used by 50K employees.
Apple will roll out Apple Intelligence in October (instead of September), and also volunteered to join the White Houseās AI Risk plan.
Treats To Try.
*Statsig powers OpenAI, Character.ai, and Anthropicās feature management and experimentation. Discover the AI experimentation playbook and test it out yourself with 2M Events Free.
Kotae provides a chatbot that automates customer inquiries for small businesses using their training files and FAQs.
Folderr enables users to create custom AI assistants and automations using their own data and third-party integrations.
Skipit summarizes YouTube videos up to 12 hours long so you can ask questions and get answers about videos you want to skim.
Animate old photos brings old photographs to life with optional custom prompts for motion effects.
Julius provides a clever UI to chat with your data, generate insights, create visualizations, and perform advanced analytics across various file formats.
*This is sponsored content. Advertise in The Neuron here.
Monday Meme
A Cat's Commentary.
Thatās all for today, for more AI treats, check out our website. Get your brand in front of 450,000+ professionals here. See you cool cats on Twitter: @nonmayorpete & @noahedelman02 |
|