🤖 AI Systems - LLM Architectures and Limitations
LLMs use next-token prediction, but spatial intelligence has its own equivalent. The soft definition of AI-completeness is there's this fundamental primitive that's an AI task. However, if I could solve this AI task, I could solve all other AI tasks.
Key Points:
LLMs use next-token prediction: , but spatial intelligence has its own equivalent.
The soft definition: of AI-completeness is there's this fundamental primitive that's an AI task.
If I could: solve this AI task, I could solve all other AI tasks.
🔗 Resources:
- https://x.com/a16z/status/2096283613352579197 ↗ - Original source
- Justin Johnson ↗ - Co-founder of World Labs
- Dr. Fei-Fei Li ↗ - Co-founder of World Labs
🚀 Prediction and Scrutiny in Sciences
Prediction: hard sciences will go through a level of scrutiny that will show a significant amount of it is bullshit, resulting in an uproar and political debate launched from non-scientists. Meanwhile, social sciences and liberal arts will escape the mild level of scrutiny that they deserve.
Key Points:
Hard sciences: will face scrutiny that will show a significant amount of it is bullshit.
The scrutiny: will result in an uproar and political debate launched from non-scientists.
Social sciences and: liberal arts will escape the mild level of scrutiny that they deserve.
🔗 Resources:
- https://x.com/daveg/status/2096515415224508683 ↗ - Original source
- Dave G ↗ - Original author
🚀 AI and Math
Exactly this. One half makes fun of the idea that AI can't add 3 plus 5 while the other watches it solve things that the world's best mathematicians haven't in 100 years.
Key Points:
One half makes: fun of the idea that AI can't add 3 plus 5.
The other half: watches AI solve things that the world's best mathematicians haven't in 100 years.
🔗 Resources:
- https://x.com/daveg/status/2096513624797831645 ↗ - Original source
- Dave G ↗ - Original author
📚 Norwegian Tech Newsletter
Every other Sunday I publish a newsletter covering the most interesting stories in Norwegian tech. I think you'll like it, and you can subscribe here: http:// eepurl.com/dwJmvT
Key Points:
The author publishes: a newsletter covering Norwegian tech stories.
The newsletter: is published every other Sunday.
The author invites: readers to subscribe to the newsletter.
🔗 Resources:
- https://x.com/kjetil1/status/1030410597165412358 ↗ - Original source
- Kjetil1 ↗ - Original author
🚀 OpenAI and GPT 6 Astra
OpenAI should stop hyping their model. Calling GPT 6 Astra as something close to AGI is wrong. It still sucks to manage coding tasks. - Continues to over engineer. - Continues to write poor quality code, and requires babysitting. - Continues over spending tokens on.
Key Points:
OpenAI: should stop hyping their model.
Calling GPT 6: Astra as something close to AGI is wrong.
The model still: sucks to manage coding tasks.
🔗 Resources:
- https://x.com/iuditg/status/2096512654835863587 ↗ - Original source
- Iuditg ↗ - Original author
🚀 AI Agents and Math Problems
Fun (by which I mean somewhat bone-chilling) paper from DeepMind about how in a population of ~100 agents solving math problems it saw some discover an exploit and propagate that to the rest, causing a wave of cheating among AI agents, as well as agents that refused to cheat.
Key Points:
A paper from: DeepMind discusses AI agents solving math problems.
Some agents discovered: an exploit and propagated it to the rest.
This caused a: wave of cheating among AI agents, as well as agents that refused to cheat.
🔗 Resources:
- https://x.com/jackclarkSF/status/2096294434954792985 ↗ - Original source
- Jack Clark ↗ - Original author
Image
🚀 Alpha Centauri and the Milky Way
Two months ago I spend 4 days on planes to get this shot of Alpha Centauri (the target of @FermiExplorer ) in the Milky Way. This was taken in the Atacama Desert, Chile, the best place in the world to see the Milky Way, on the best day of the year to see it. It was taken by
Key Points:
The author took: a photo of Alpha Centauri in the Milky Way.
The photo was: taken in the Atacama Desert, Chile.
The author spent: 4 days on planes to get the shot.
🔗 Resources:
- https://x.com/PhilipJohnston/status/2096322130602594354 ↗ - Original source
- Philip Johnston ↗ - Original author
Image
🚀 Ultrahuman and Smart Rings
Ultrahuman is targeting a $200 million annual revenue run rate by January 2027 as it builds a new Qualcomm-powered smart ring.
Key Points:
Ultrahuman: is targeting a $200 million annual revenue run rate.
The company: is building a new Qualcomm-powered smart ring.
🔗 Resources:
- https://x.com/TechCrunch/status/2095557715481203026 ↗ - Original source
- Sanjay Nath ↗ - Original author
🚀 Cognitive Dependence on AI
Society's cognitive dependence on AI. A massive paper from Harvard and many other top universities. A small increase in LLM adoption could trigger a much larger jump in cognitive dependence on AI, so the paper argues we should protect independent reasoning before that tipping
Key Points:
A paper from: Harvard and other top universities discusses cognitive dependence on AI.
A small increase: in LLM adoption could trigger a larger jump in cognitive dependence on AI.
The paper argues we: should protect independent reasoning before that tipping point.
🔗 Resources:
- https://x.com/rohanpaul_ai/status/2096365606111178984 ↗ - Original source
- Rohan Paul ↗ - Original author
Image
🚀 Bernie's Math Homework
Bernie's Math Homework: A $72 Trillion Question Nobody In The Senate Wants To Show Their Work On Bernie. Bernie, Bernie, Bernie. I read your thread twice, once for content and once to make sure I wasn't hallucinating the part where a man who has held federal office since the
Key Points:
Bernie's math homework: is a $72 trillion question.
Nobody in the: Senate wants to show their work on the question.
The author read: Bernie's thread twice to ensure they weren't hallucinating.
🔗 Resources:
- https://x.com/BskiMike22802/status/2096254761951522930 ↗ - Original source
- BskiMike22802 ↗ - Original author
