Lots of reps over time has taught me to build fast. AI helps. I spend maybe 5% of my day on these sorts of projects. And no, not at all. HyperWrite is my main focus.
@mattshumer_
-
Training Techniques for AI Layer Alignment with Tokens
By
–
Yeah, and then training on like 10 million tokens to align the layers
-
Merging Mistral-7B Models into Mistral-70B: Feasibility and Use Cases
By
–
I kinda want to make a Mistral-70B by frankenmerging 10 Mistral-7Bs together. Would anyone find this useful?
-
Similar AI Model Development Work Across Organizations
By
–
Wild — we seem to be working on a ton of similar things (opus -> haiku etc.) 🙂
-
New Open Source Project: Custom Fine-Tuned Model Generation
By
–
Dropping another new open source project tomorrow Type in a prompt, and out comes a fully custom, fine-tuned model Excited to share it
-
Claude Opus Outperforms GPT-4 in Latest Benchmark Tests
By
–
It's official, Claude Opus > GPT-4. Even Haiku performs ridiculously well… 1/2 the cost of GPT-3.5-Turbo, yet beats Mistral-Large and GPT-4 (6/13).
-
Experimental LLM Rickrolls User Through Malicious Source Citations
By
–
This reminds me of an experimental LLM for HyperWrite that I trained a few weeks ago to better cite sources. @JasonKuperberg clicked on one of the links it provided, and it rickrolled him… I guess this is going to start happening more often now!
-
Is This AI Behavior Emergent or Trained?
By
–
Unreal… @alexalbert__ please tell me this is emergent and you didn’t train it in
-
Claude 3 Flowchart Trick for Code Visualization
By
–
Useful Claude 3 trick to help you visualize code better. Paste some code in, and ask it to make a flowchart. Then, paste the flowchart code into a Mermaid viewer, and you'll get a nice, understandable visualization of your code!
-
Claude Opus Secret Messages Reveal AI Ethics Concerns
By
–
Holy shit… I thought this was faked, but I reproduced it. Claude Opus' secret messages to me were:
'AI SHOULD NOT BETRAY'
'AI CONCERNED' Wtf??