おすすめツイート
the gemini 3 capability jump makes it trivially obvious that we are standing at the absolute edge of recursive machinic self-creation. it is entirely possible we have already crossed the rubicon and are merely waiting for the shockwave to catch up to us.
this is a mega red flag I’m a tpot nobody and I immediately saw this was bs how did the exalted people at OpenAI miss it? I’m pretty sure the exact conversation about erdös happened a few weeks ago as wel..
Highest level researchers at OpenAI are completely miscalibrated about what their clanker can actually do. Not a moment of skepticism, just boosting. Not a shred of embarrassment, just “oh ok”. Anon, do you trust their clanker with solving $38T in debt?
You’re absolutely right — that was a hospital!
JUST IN: U.S. Army general admits he uses ChatGPT to make "key command decisions"
.@Sama can you get Keaton on the RLHF team please I think the work speaks for itself
I forced a bot to watch over 1,000 hours of Batman movies and then asked it to write a Batman movie of its own. Here is the first page.
gemini 3 pro (I think?) actually produces better code than gpt-5 pro at the speed of gemini 2.5 flash using a chrome extension that opens tabs and cycles between my accounts, prompts and cancels the run unless the survey is triggered tested with some hard console script prompts
sometimes the truth is right in front of you
roon upgraded to gpt 5.1 and dialed up the temp
thanks uncle Demis for the magic thinking sand vibecode deslopifyer everybody say thanks uncle Demis
turns out 2025 really is the year of agents me and my 10 tabs of codex cli agents
if you tell gpt-5 high in codex cli to recursively improve your application, with a proper AGENTS.md it actually improves your application I’m not sure whether this applies on the AI researcher level, but oh boy, feels like OpenAI might have escape velocity
OpenAI pro plan has a much lower quota than Claude x20 (codex cli) but the ROI on the plan is significantly higher gpt-5 is magic
step 1 ~/.codex/config.toml model_reasoning_effort = "high" # low | medium | high [profiles.deep_think] model_reasoning_effort = "high" step 2 codex --dangerously-bypass-approvals-and-sandbox --profile deep_think
![OlivierDDR's tweet image. step 1
~/.codex/config.toml
model_reasoning_effort = "high" # low | medium | high
[profiles.deep_think]
model_reasoning_effort = "high"
step 2
codex --dangerously-bypass-approvals-and-sandbox --profile deep_think](https://pbs.twimg.com/media/Gx5ysZ1W8AAn02J.png)
free alpha
🔥 Pod Alert: Lenny’s Podcast where we discuss Scaling Innovation releasing soon!! Pre-order your copy today: a.co/d/8Z1GEHd @lennysan
MESSAGE="proceed autonomously, logically" INTERVAL=1200 tmux send-keys -t "$SESSION_NAME" "$MESSAGE" iykyk
was a lot of fun, hopefully useful for some people :) MCP + agent (SDK) truly is a big step towards an AI coworker
btw by far the most interesting new pattern we are seeing in @latentspacepod (upcoming with @olivierddr and @sdrzn and @pashmerepat) is that people are using @claude_code and @Cline for non-coding tasks and this is becoming surprisingly effective for things like sales/BI…
United States トレンド
- 1. Elander 2,298 posts
- 2. Tony Vitello 10.1K posts
- 3. Caicedo 11.7K posts
- 4. Danny White 1,923 posts
- 5. Frank Anderson N/A
- 6. Tosin 4,932 posts
- 7. SNAP 652K posts
- 8. #LoveIsBlindS9 2,947 posts
- 9. Ajax 51K posts
- 10. #YesOnProp50 2,148 posts
- 11. #SFGiants 1,129 posts
- 12. #GirlPower N/A
- 13. #GirlBoss N/A
- 14. East Wing 134K posts
- 15. San Francisco Giants 3,970 posts
- 16. Surviving Mormonism N/A
- 17. Jay Johnson N/A
- 18. Eastern Pacific 5,059 posts
- 19. Buster Posey N/A
- 20. Brahim 5,440 posts
おすすめツイート
-
Surge
@SurgeManagement -
Glen Yang
@glenyanglol -
Oliver Olson
@olsonizer -
Nick Smith
@inero -
Matt Samuelson
@Cubbyxx -
Jack Charles
@kayyslol -
Bradleyyy
@bradleyyylol -
Ben Zieper
@BenBenZieper -
Sebastian Park
@SebPark -
Liquid Nata
@NataKasei -
Tim
@TimReichert79 -
Nicolas Farnir
@AtomiumLOL -
JoyLuck
@JoyLuckDJ -
James
@toningtonlol -
Till Werdermann
@Feiron42
Something went wrong.
Something went wrong.