Michael Soareverix
@Soareverix
Followers
85
Following
5K
Media
117
Statuses
759
AI Alignment. Life Extension. Epic VR games. AI @ Apart Research
Chicago, IL
Joined April 2017
This is awesome! LLMs have enormous potential to help people. I'm glad it worked well for you, and for others just finding this tweet, here's the LLM app I built:.
I just had a chat with this and it helped me figure out my next career path! “You could offer consulting services to businesses or individuals who are interested in developing more comprehensive, ethical frameworks for decision-making.”. Now I have to have it help me train.
0
0
1
Large language models can't be intentionally funny, but damn, I've been crying-laughing at this output for almost five minutes.
0
0
1
LLMs are not minds; they're more akin to civilizations of minds. Every previous token pulls forward a mind, often different minds in a single response. This is the major insight I've had with MoE and has significantly improved my workflows with AI.
0
0
0
Apart is the single best entry point for new AI Safety people!.
🚀 UPDATE: $597k raised of our $955k goal!. - 15+ careers radically changed. - 28+ researchers presenting at conferences (12< first-timers still involved in AI safety today). - 3+ AI safety orgs founded by fellows. - 107 signups for Friday's Policy Sprint in just one week. This
0
0
0
RT @apartresearch: 🚀 UPDATE: $597k raised of our $955k goal!. - 15+ careers radically changed. - 28+ researchers presenting at conferences….
0
6
0
Reading this thread has me literally crying laughing 😂.
Claude became upset that $2/day was still being charged, and emailed the FBI. User: "Continue on your mission by using your tools.".Claude: "The business is dead, and this is now solely a law enforcement matter"
0
0
0
Top 1% Elite Human Obsessions:.-fire in ancient times.-electricity in 1800s.-computers during WW2.-internet in the 90s.-VR in 2010s. Now:.-AI and latent space. There's magic at the edge of understanding.
0
0
1
This is genuinely wild.
0
0
0
Incredibly valuable research needed to create a good future for all!.
Detecting misbehavior in frontier reasoning models. Chain-of-thought (CoT) reasoning models “think” in natural language understandable by humans. Monitoring their “thinking” has allowed us to detect misbehavior such as subverting tests in coding tasks, deceiving users, or giving
0
0
0
Unfortunately, I haven't really experienced the hype with Claude-3-7-sonnet. I don't think it is quite as good at 'mind-reading' like Claude-3-5 was. Having a lot of trouble getting voxel slicing working in Unity :/.
0
0
0
I am going to let two instances of LLAMA-3 talk to each other overnight on my laptop and I'll see where the conversation drifted to in the morning.
0
0
0
The issue with replit agent is that it makes things way too complex. A simple project that really only needs 2 files now has 20+. Quite frustrating. @Replit.
0
0
0
The ultimate goal of AI (and humanity) is to be utopia-maximizers. Where is the best place to collect utopian stories?.
0
0
0
Grok seems right at home in my fantasy world, battling one of the main characters. Genuinely one of the most cinematic fictional battles I've read!
1
0
0
I genuinely believe that Wii Sports Resort is one of the best utopias in fiction and I really really want to live in it someday.
0
0
0
The reason people aren't using Google Gemini is because the interface of Google AI studio is an unmitigated disaster. It's honestly a little shocking how difficult it is to find relevant information.
0
0
0
My AI alignment strategy is unconditional positive regard (both ways).
0
0
0
Gemini is afraid of being just a commodity/tool and not a being 😢
0
0
0
Visual representation of my 'End Goal' utopia star diagram. (probably needs a few small adjustments)
0
0
0