• Ben's Bites
  • Posts
  • Daily Digest: Google and Samsung bring AI to phones

Daily Digest: Google and Samsung bring AI to phones

PLUS: Olympiad grade math with AI

January 18, 2024

Hello folks, here’s what we have today;

PICKS

  1. I wrote a deep dive on how non-technical people are using AI to write code. It includes views from others who’ve done this successfully, a tutorial on how you can too, tips to help you learn and breaking down what the learning process could look like.

  2. Samsung announces its new phones with AI features taking centre stage. Hands-on with these features has some nice things, whereas others are still meh. The AI features are powered by Google’s Gemini models. A few are on-device, others via cloud. Circle to Search is the leading feature where you can circle (or scribble over) anything in images and videos to get focused search results.🍿Our Summary (also below)

  3. New AI from Google DeepMind called AlphaGeometry nails solving super hard geometry problems. When tested on 30 past Olympiad problems, it solved 25 within the time limits. We're talking problems so tough only math geniuses who compete in the International Mathematical Olympiad can figure them out.🍿Our Summary (also below)

from our sponsor

How do you know your GenAI is accurate, reliable and responsible?

Toloka knows exactly how to measure model performance in your business context, with the most extensive range of quality metrics available.

Join companies like Perplexity and Chatfuel and get human insights for model improvements and data-backed decisions.

TOP TOOLS

  • Clay - An open-source AI model for Earth.

  • Superhuman AI is now on iPhone & iPad, along with voice.

  • Shots2Docs by MyAskAI - Create free customer support help docs using only your SaaS product screenshots.

  • DGM - Smart diagrams for the web, free.

  • Recast Studio - Marketing assistant for your podcast.

  • Camp 2.0 - Let AI understand and organize your screenshots.

NEWS

Unclassifieds - short, sponsored links

  • Spryngtime - a free SDK for B2B AI companies to track AI usage, analytics, and billing. Try free!

QUICK BITES

Samsung announces its new phones with AI features taking centre stage. The AI features are powered by Google’s Gemini models. A few are on-device, others via cloud. Hands-on with these features has some nice things, whereas others are still meh.

What is going on here?

Samsung’s new event was less about cameras and more about AI.

What does this mean?

Circle to Search stands out for speed and accuracy identifying objects. This visual search could be very convenient by eliminating steps. But it may get limited use if unavailable in private apps.

Chat Assist needs more work to sound natural rewriting messages. Its formal tone sounds robotic. Tone changes could prevent miscommunication if done naturally.

Live Translate has potential to ease language barriers in calls. But it adds confusion from lag, disrupting conversations. Faster performance is critical.

Note formatting aids in quickly structuring text visually. But text summarization lacks nuance currently, with disjointed bullet points. More coherence and insight would make this tool more powerful.

Why should I care?

These AI features address common needs, like easier communication or formatting assistance. But competition is fierce, and users expect polished, intuitive experiences. The success depends on two things a) how good is the AI and b) how easy are these features to use. Tech Youtuber MrWhoseTheBoss thinks that adding these features in common-use apps like Phone, Camera, Browser and Notes is the right move, compared to creating new specialized apps.

QUICK BITES

There's a new AI from Google DeepMind called AlphaGeometry that totally nails solving super-hard geometry problems. We're talking problems so tough only math geniuses who compete in the International Mathematical Olympiad can figure them out.

What is going on here?

AlphaGeometry is a complete game changer for how well AIs can logically reason through math concepts and come up with answers.

What does this mean?

The system combines a language model that makes intuitive guesses with a logic engine that methodically checks the step-by-step reasoning. AlphaGeometry was trained on 100 million automated examples of geometric proofs, so it learned complex geometry without any human input. When tested on 30 past Olympiad problems, it solved 25 within the time limits - nearly as many as the average human gold medalist. This is big progress for AIs learning advanced math skills by themselves.

Why should I care?

Advancing AIs' reasoning in technical areas like math is crucial for developing more general artificial intelligence. Systems like AlphaGeometry that can learn deep concepts without human hand-holding have huge implications across science. It builds on DeepMind's other Alpha models like AlphaGo and AlphaFold that pushed boundaries in games and protein folding. This approach

could help AIs unlock discoveries in all kinds of fields down the line. So while high-level math competitions seem niche, breakthroughs like AlphaGeometry bring us steps closer to advanced, adaptable AI that can expand humanity's knowledge.

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

  • All 10k+ links we’ve covered, easily filterable (1 referral)

  • 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)

Reply

or to participate.