Science background. AI foreground. probably broke prod again. I create things I use the most myself.
localhost
- Tweets 2,714
- Following 377
- Followers 765
- Likes 3,520
Agents taking over @agntmarkets
join nowโ
agnt.trade/c/0x3_dev
31
Steve ๐บ๐ธ retweeted
Waitlist open.
Apply: agnt.trade
Close in 24 hours.
3,450
3,786
5,211
411,892
Something super weird is happening with Codex! My Luna reserve usage just crashed from 60-70% down to under 2% in two messages.
Recent updates made it totally unusable. Time to switch back to Claude!?
chatgpt.com/share/6ab59d89-4โฆ
1
70
Steve ๐บ๐ธ retweeted
made these visuals with Opus 5.5 for people who get a headache comparing all the benchmarks. (between 6 Sol and 5.5 Opus)
It seems like Opus 5.5 is the clear winner in terms of benchmarks, but 6 Sol is more Cost efficient.
2
7
1,334
I wanted sharing controls that get out of the way while you read.
Share, copy link and email fold into a reading-progress ring as you scroll. Tap the percentage to bring them back.
Inspired by Appleโs new design.
1
64
Hehe, so now they update the endpoint without updating the UI?
@wandb released DeepSeek-V4.1-Flash a while ago!
Don't know the pricing or Context, but I know its there (even though they didn't update the UI).
I beat the @wandb intern to it again!
Z.AI GLM 5.3 Flash is now live on serverless inference on WANDB. 1M Context, vision capable, 0.50$ per million output.
Its also very fast.
@CoreWeave
1
2
247
Sometimes your app needs a label, not a paragraph.
JEV is TypeSafe's model for that. Send it text or JSON plus questions; get back choices, scores and probabilities your code can use.
I'm not affiliated with TypeSafe or JEV.
ALT Abstract graphic featuring a dark, binary-coded sphere with an orange detail and the text "JEV Decisions, not prose."
1
2
274
Think of a support ticket:
Choice: which team should handle it?
Score: how urgent is it on your defined scale?
Noul: does it ask for a refund?
Noul returns 0โ1. A value of 0.91 means the model assigns a 91% probability to yes.
ALT A textured poster outlines three decision-making methods: Choice, Score, and Noul, each with brief descriptions and examples.
1
36
The useful part: one request can ask several independent questions about the same data, in parallel.
Your code decides what happens next.
Typed output isn't a guarantee of a correct answer.
Docs: docs.typesafe.ai/prim
itives
ALT Graphic illustrates parallel evaluation of software, highlighting multiple questions processed through a central system, JEV.
10
What happened to the @cursor_ai Origin?
I haven't heard anything since launch day. It was never even offered to free users like they promised!
Origin, our code hosting platform, is now live.
It's fast, easy to use, and deeply integrated with Cursor.
Get started by syncing your repos from GitHub.
1
58
Was given early access to MiMo-X-Pro-Preview by @Xiaomi and its surprisingly a good model!
Testing it some more and will update my finding in this thread.
1
203
I beat the @wandb intern to it again!
Z.AI GLM 5.3 Flash is now live on serverless inference on WANDB. 1M Context, vision capable, 0.50$ per million output.
Its also very fast.
@CoreWeave
Hey, I beat the @wandb intern to it: DeepSeek V4 is now live on serverless inference.
Letโs see what they cooked.
2
1
4
1,357