sunilpaul.bsky.social
@sunilpaul.bsky.social
Reposted
For those of us who always stressed that "AGI" was an empty concept, what's happening now looks like

1. Substantive utility claims for AI are broadly acknowledged.
2. "But it's not AGI" == a politic gesture that allows critics to save face.

I'm not criticizing Chris; this is how we move forward!
thiiiiis

semantic search is CRAZY useful in how it can unlock the value of the vast majority of data, which is unstructured

the fact that LLMs can also summarize and recontextualize that unstructured data is truly a breakthrough

this is not by any stretch of the imagination AGI
A good 70% of the fighting over AI would have been avoided if VC hucksters (and gullible access journalists) hadn't conflated incrementally useful automation with computational sentience to make money
July 11, 2025 at 8:30 PM
This is a critical question
It seems like a big question in AI right now is whether there is a flywheel in AI development & when it kicks in, either because having advanced AIs let you code & train better AIs, or due to other returns to experience in building AI

If/when it does, all non-frontier labs fall behind the race.
April 19, 2025 at 2:43 AM
Reposted
Here is the full text of Rachel Cohen’s firm-wide email:
March 21, 2025 at 1:02 PM
This just in: Musk sues European consumers insisting they must buy Teslas. electrek.co/2025/02/05/t...
Tesla sales dropped 60% in Germany
Tesla’s sales have dropped nearly 60% in January in Germany compared to the same period last year. The same thing...
electrek.co
February 22, 2025 at 1:14 AM
Today Ronald Regan is turning in his grave as Putin & Trump slow dance over it together.
February 21, 2025 at 1:00 AM
@emollick.bsky.social - I appreciate that you are posting all your material here on Blue Sky, now. I followed you on Twitter from before chatGPT 3.5 and enjoyed the social science posts. Now am loving your AI focus and still including occasional social science/innovation/entrepreneurship posts.
February 9, 2025 at 9:00 PM
The emergence behavior of neural nets continues to astound. Deepseek also had a (very different) approach that also involved fewer GPUs but running more cycles. So did Apple with its AV example (see prior repost by me from @emollick.bsky.social ).
This paper is wild - a Stanford team shows the simplest way to make an open LLM into a reasoning model

They used just 1,000 carefully curated reasoning examples & a trick where if the model tries to stop thinking, they append "Wait" to force it to continue. Near o1 at math. arxiv.org/pdf/2501.19393
February 7, 2025 at 5:44 AM
Reposted
Apple team shows self-driving AI can learn entirely by practicing against itself - no human driving data needed

In testing, their system averages 17.5 years of continuous driving between incidents, far surpassing humans. All through self-play, not imitation.​​​​​​​​​​​​​​​​ arxiv.org/abs/2502.03349
February 7, 2025 at 1:20 AM
🚨 Elon Musk’s DOGE: Efficiency or Empire-Building? 🚨

Musk runs Department of Government Efficiency (DOGE) while 60% of his fortune is tied to federal funding, contracts, and regulation.

Is Musk fixing Washington—or fixing the game? 👇
February 7, 2025 at 4:49 AM
Why isn't this being mentioned in all the press about DeepSeek?
January 30, 2025 at 12:49 AM
Trump's first acts as president:

* pardoned criminals who attacked police
* enriched himself by launching a cryptocoin
* denied disaster aid for fellow Americans who lost their homes

We won't forget.
January 24, 2025 at 7:29 PM
Published in 2017 by @rodneyabrooks.bsky.social , this article about AI is worth re-reading in today's AI hype. Favorite quotes:
- We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.
- Nothing in the universe is without limit.
The Seven Deadly Sins of AI Predictions
Mistaken extrapolations, limited imagination, and other common mistakes that distract us from thinking more productively about the future.
www.technologyreview.com
January 22, 2025 at 1:01 AM