Instinct is next level, but most important of all - a very effective "Weight Watchers"! @noahrshinn, @instinct
49
Want to use Codex / GPT-6.1, but it is painfully slow. Half the time, just sitting there waiting for the agents to finish. @thsottiaux (Tibo)
1
98
Tibo @thsottiaux - Is Codex running in slow motion today for anyone else?
/fast feels like a polite suggestion at this point. What is it even doing? 🐌
1
120
Tibo - In the Codex macOS app, does enabling /fast mid-turn take effect immediately, on the next model request after a tool call, or only when a new turn starts? @thsottiaux
Useful to know for balancing speed and weekly quota during long-running tasks.
137
Hello Tibo (@thsottiaux)- Small feature/bug request. Can codex put thread titles in "Menu->Window" listing, instead of "ChatGPT" for all windows? Thx.
123
I am experiencing very slow performance on codex macos app (pro max plan) with 6-luna-max?
Anyone else seeing the same?
Tibo @thsottiaux - fyi. In case you are aware of it, I presume openAI is reviewing it. If not....
This (kind of linear with log-scale cost) intelligence/cost plot makes it easier to choose depending upon task classification and expense appetite !!
GPT6 Luna
GPT6 Sol
Opus 5.5 (replaces Fable 5.1)
GPT6 Astra (6.1 on the way?)
Fire around ruby hills Pleasanton. Stay safe.
share.watchduty.org/i/110154…
#LittleFire
Plane dropping fire retardant on hills affected by the Little Fire near Ruby Hills in Pleasanton (~1000 acres burnt; only 10% contained)
share.watchduty.org/i/110154…
#LittleFire
3
2,638
Dev Vidhani retweeted
Sweat when you see math in an ML paper 🤨?
Relax - I'm here to break it down for you.
We'll start with RoPE, Llama 2's positional embedding scheme 🍃 👇 🍃
Dev Vidhani retweeted
The human brain is far to good for the purpose it has evolved. Intelligence might therefore suddenly emerge through some kind of phase transition at some level of complexity. Might something similar have happened to GPT-4? I find it too intelligent for how it is trained.
In addition to Mustafa’s proposed ACI Turing test, which offers a commercially driven approach to bridge the gap between AI and AGI, a new test should present a longer-term, more ambitious objective while maintaining simplicity and clarity. #AI #TuringTest #Opinions #AGI #ACI
AIs today are close to mastering language... We need a new milestone - a Modern Turing Test - to measure progress for the next phase. Once they can take actions, we have artificial capable intelligence (ACI) and the full transition to a post-AI world begins.
What matters is what a system can actually do, its capabilities. One objective measure for this would be to see if, with no real intervention, it can make $1 million on the Internet. This is a reliable proxy for a whole raft of capabilities.
Should this Modern Turing Test be met, it means we are in a new phase of AI. Agents will have substantial real world impact, not just on the economy but likely across every area of human life. They will be everywhere, and they will be DOING things.
Read more in my new article: technologyreview.com/2023/07…
1
294
Therefore, an AI should be programmed to prioritize collective benefits over individual gains, requiring complex decision-making, responsibility, and foresight in its actions.
1
37
Thus, co-survival skills could serve as a more comprehensive and nuanced evaluation metric for AI. They indicate not just an AI's cognitive abilities, but also its capacity for ethical reasoning, social collaboration, and long-term sustainability.
#AI #TuringTest #AGI #ACI
30





