prabal97

5 ms·

1.
GPT-6 Astra Is Here–and OpenAI Thinks It May Kick Off the AGI Era(wired.com)
5 points by prabal97 13d ago|0 comments
2.
Ab(use) of LLMs for Learning(prabal.ca)
1 points by prabal97 15d ago|0 comments
3.
by prabal97 16d ago
I mean an agent probably only uses one mac at a time. But they're obviously training tons of agents concurrently.
4.
by prabal97 16d ago
I think they just want Apple devices to train their computer-use models in a realistic environment.
5.
Apple Is Suddenly an AI Infra Stock as OpenAI Buys 10k+ Macs(247wallst.com)
41 points by prabal97 16d ago|24 comments
6.
AI and the Death of Expertise: Why Experts Still Matter(inc.com)
3 points by prabal97 17d ago|0 comments
7.
New AI Fear Unlocked as Traders Ditch Companies at Risk of Disruption(bloomberg.com)
5 points by prabal97 20d ago|0 comments
8.
India's campus talent landscape pivots to skill-based hiring: Deloitte India(deloitte.com)
2 points by prabal97 21d ago|0 comments
9.
AI reshapes India's IT services sector contracts as clients demand more for less(reuters.com)
3 points by prabal97 24d ago|0 comments
10.
Scientific spat erupts over claim that bones have key immune vessels(science.org)
3 points by prabal97 27d ago|0 comments
11.
I had to unplug from AI to rediscover my love of writing(science.org)
2 points by prabal97 28d ago|0 comments
12.
Landmark cancer trial shows success against an 'undruggable' cancer(nature.com)
2 points by prabal97 2mo ago|0 comments
13.
A GPT-5.4 bug led to OpenAI banning goblins and raccoons
6 points by prabal97 5mo ago|1 comments
14.
by prabal97 5mo ago
You can use stuff like Litestream to continuously back up your SQLite DB to a cheap storage bucket. It's practically set and forget too.
15.
Using Claude Code with my ChatGPT subscription instead of paying for both(prabal.ca)
5 points by prabal97 5mo ago|0 comments
16.
by prabal97 5mo ago
FYI I use my Codex models with Claude code and they work pretty great. It can even pick up on existing conversations w/ Opus and then resume w/ OAI models.
17.
by prabal97 5mo ago
I use my OAI subscription on my Claude Code. I get the benefit of the Claude Code interface with the intelligence of OAI models. https://prabal.ca/posts/claude-code-chatgpt-subscription/
18.
DNA polymerases can generate long, complex sequences without templates(nature.com)
2 points by prabal97 5mo ago|0 comments
19.
'Treasure trove' of antiviral proteins could inspire powerful molecular tools(nature.com)
6 points by prabal97 5mo ago|0 comments
20.
Half of social-science studies fail replication test in years-long project(nature.com)
3 points by prabal97 6mo ago|0 comments
21.
Chip-scale optical wireless system hits 362 Gbps at half the energy of Wi-Fi(spiedigitallibrary.org)
10 points by prabal97 6mo ago|0 comments
22.
by prabal97 6mo ago
> in a single working day those trucks would shift rocks equivalent to the weight of the Empire State Building. Oh. My. God.
23.
Scientists crack a 20-year nuclear mystery behind the creation of gold(sciencedaily.com)
87 points by prabal97 6mo ago|39 comments
24.
ClaudeDown: Is Claude getting dumber, or is it just you?(claudedown.com)
3 points by prabal97 6mo ago|2 comments
25.
Bourbon waste could provide next-gen supercapacitor components(spectrum.ieee.org)
38 points by prabal97 6mo ago|15 comments
26.
The Revenge of the Data Scientist(hamel.dev)
1 points by prabal97 6mo ago|0 comments
27.
by prabal97 6mo ago
Reposting it here ... I wrote this more intuitive explanation. I think you might find it helpful too! https://prabal.ca/posts/google-long-context-cheaper/
28.
by prabal97 6mo ago
I wrote this more intuitive explanation. I think you might find it helpful! https://prabal.ca/posts/google-long-context-cheaper/
29.
by prabal97 7mo ago
They're weighted differently. Currently it's set to: reposts 2.0x, replies 1.5x, bookmarks 1.2x, likes 1.0x, and clicks 0.5x. These are guestimates based on public info.
30.
by prabal97 7mo ago
It's real embeddings (MiniLM, truncated to 128-dim), so the cosine similarities between tweets are genuine, but it's obviously not X's actual trained model. The relative distances are meaningful, though.

More ›