Jacques
@JacquesThibs
Followers
4K
Following
23K
Media
2K
Statuses
15K
Automating R&D safely and securing the future. šØš¦ Building something new.
San Francisco, CA
Joined May 2008
From a Paul Christiano talk. Assume we're in a world where AI systems are broadly deployed and the world has become increasingly complex, where humans know less and less how things work. A viable strategy for AI takeover is to wait until there is certainty of success. If a 'bad.
11
4
80
If Iād have a company with multiple offices, I would add a big monitor that is always on in the lunch area so that it acts as a two-way portal to see and talk to each other.
0
0
1
fyi, this does not mean weāve āsolved mathā.
Most important part of the IMO Gold achievement. Were you surprised by this? Did you not update all the way to avoid likelihood of surprise?.
1
0
4
As a side note, we may be seeing signs of a frightening future. One can hope such an approach just stays safe and useful until we can use it to take a safer path to ASI.
1
1
5
Most important part of the IMO Gold achievement. Were you surprised by this? Did you not update all the way to avoid likelihood of surprise?.
So whatās different? We developed new techniques that make LLMs a lot better at hard-to-verify tasks. IMO problems were the perfect challenge for this: proofs are pages long and take experts hours to grade. Compare that to AIME, where answers are simply an integer from 0 to 999.
3
1
14
Oh, and you also didnāt create a notification sound for when one of your agents have finished their task and need your assistance.
0
0
0
If you find yourself going on social media too much while coding with AI and waiting for generations, you are simply not using enough AIs. Personally, Iām almost never waiting around and instead AIs are waiting around for me to give them further direction.
2
0
19
?.
Today the OpenAI Nonprofit Commission released a report asserting OpenAI is āpositioned to be a force of good in this Age of Intelligenceā. The report also claims that AI Now was among the orgs participating in the listening process that led to this report. We did not.
0
0
2
Running Multiple Claude Code Accounts. Add these lines to your ~/.zshrc or ~/.bashrc file to create shortcuts for different accounts. The magic is the CLAUDE_CONFIG_DIR environment variable. It tells Claude where to store its config, keeping accounts completely separate:. alias.
1
0
1
Anyone know what's the best way to easily manage multiple Claude Max accounts on the same machine? I don't want to have to log in and out to access both, but still want to use my local environment.
2
0
1
Great thread from someone who got a *speedup* from using AIs in the METR study. He points to many things I try to reinforce in people when giving a workshop on using AIs in research.
I was one of the 16 devs in this study. I wanted to speak on my opinions about the causes and mitigation strategies for dev slowdown. I'll say as a "why listen to you?" hook that I experienced a -38% AI-speedup on my assigned issues. I think transparency helps the community.
0
0
0
When you try to actually build with AI, I think you also gain an understanding of AI capabilities/progress that you might miss if you only read AI safety literature
I think you gain a special level of insight for how fast the world is going to change if you are actively trying to leverage AI to build new businesses as fast as possible.
1
0
3
ok, I'm not the only one. Claude Code seems to be stopping without any error message. At most it is reading a file and then just stops.
3
0
7