s3tt3mbr1n1

Thanks, I feel seen.

There is no Antimemetics Division…

That’s true, but as their model quality is behind Kimi K2.6 or Deepseek v4, I believe it makes more sense to consume those Chinese models from a European hosting company like Scaleway, than to consume Mistral’s models hosted on US cloud providers, except if you see that Mistral subscription as a kind of investment into Mistral for them to buy infra in the future. That is, if your goal is to “consume” AI on European soil.

I think it’s important to note that, at least for now, Mistral uses the big American cloud providers for inference [0].

[0] https://trust.mistral.ai/subprocessors

Very interested to hear more. Do you live in an (old) city or more rurally?

Replying to see if anyone has a foolproof solution.

My next attempt will include better, windmill milled flour.

Original title (too long for submission): “A small polymerase ribozyme that can synthesize itself and its complementary strand“

Abstract: The emergence of a chemical system capable of self-replication and evolution is a critical event in the origin of life. RNA polymerase ribozymes can replicate RNA, but their large size and structural complexity impede self-replication and preclude their spontaneous emergence. Here we describe QT45: a 45-nucleotide polymerase ribozyme, discovered from random sequence pools, that catalyzes general RNA-templated RNA synthesis using trinucleotide triphosphate (triplet) substrates in mildly alkaline eutectic ice. QT45 can synthesize both its complementary strand using a random triplet pool at 94.1% per-nucleotide fidelity, and a copy of itself using defined substrates, both with yields of ~0.2% in 72 days. The discovery of polymerase activity in a small RNA motif suggests that polymerase ribozymes are more abundant in RNA sequence space than previously thought.

This is sick, great job!

Smart move by Apple. I thoroughly enjoyed the first season of Severance and can recommend it to everyone.

First, thank you for working on this.

Second, a question. Computer Use and JSON mode are great for creating a quasi-API for legacy software which offers no integration possibilities. Can MCP better help with legacy software interactions, and if so, in what ways?

Has anyone gotten this to work?

Copying the repo and downloading the models through HuggingFace or manually does not seem to work, you get errors indicating missing files.

For this reason I read Andrew Yang’s “The War on Normal People”. Besides UBI and “social credits”, I don’t see him offer that many other solutions to this problem. UBI also still needs to be proven as far as I’m aware.

When o1 was released, I ran an internal eval and saw it plainly outperforming our highly educated colleagues. I had goosebumps, and haven’t been able to sleep well for days. This will dramatically impact society in 2-5 years.

Do you know of any relevant material related to this?

Interesting that this seems to be the announcement of Anthropic's Copilot alternative:

> A private beta of the Claude 3.5 Sonnet's new Fast Edit Mode, optimized for text editing. This upcoming mode achieves unprecedented speed in transforming existing text, enabling near-instantaneous code refactoring and document editing at scale.

Sounds like the Golden Mean, or Aristotle's moral virtues. [0]

[0] https://en.wikipedia.org/wiki/Golden_mean_(philosophy)#Arist...

Anything you would recommend, specifically for salary negotiations?

That’s very insightful and the argument I would expect if I were to start discussions about salary. How would you counter this?

I would also say I’m more qualified for the job than those that are applying now, so I’m not sure if this really is applicable to me.

I’m in the unique situation where I’m hiring my replacement after having made a promotion. I know they will make (substantially) more than me. How would I negotiate a matching or superior salary?

I feel like this might explain (and validate?) his conservative stance towards generative LLMs leading to AGI/ASI. What do you think?

It’s not recommended to finetune models on content that was not in the training set.

Finetuning should be reserved for desired behavior of the LLM, not to add new knowledge. For that, RAG is better suited.