DrSCM

@trajektoriePL 1- Mistral is the only recognised AI lab in the EU. 2- The EU has the habit of giving money to large conglomerates instead of risking with smaller disruptors.

2023: The European Union is desperate for its own artificial intelligence giant. German startup Aleph Alpha might be its best hope. 2026: Canada's AI startup Cohere buys Germany's Aleph Alpha to expand in Europe. Turns out "digital sovereignty" in EU was just a buzzword.

Just wrapped up a talk at the Swiss Proteomics Meeting by LS2 on integrating #multiomics data using a multi-layer generalization of #WGCNA, powered by AI. Loved the discussions with the community. Thank you to LS2 for a great event and to everyone who stopped by to chat! ✨

If you are interested in agentic bioinformatics, come and see. We take your multi-omics data, find co-regulated modules across proteomics + metabolomics + lipidomics simultaneously, and then has our agent walking you through it. #agent #multiomics #bioinformatics

@e73_da @NousResearch @Teknium Can you low-ball how much it has cost you for the session?

K2.6 + hermes = 4 hr session setting up qwen 3.6 training regime on dgx spark with the current autnomous session currently lasting 70+ min without any prompting. Kimi with hermes is next level. @NousResearch @Teknium

Friendly reminder, multi-omics is not independent omics analysis, but integrated omics.

@thorstenball @stuli1989 @AmpCode So, if main AI labs are pushing so hard on coding agents, where do you see Amp in the next 6 months? I have been a pretty much follower of Amp since the times of SourceGraph-Cody, but I see the lab pricing very difficult to beat.

@stuli1989 @AmpCode Yes, pretty much.

So this is what @AmpCode was doing all along with the Oracle and Librarian right?

@Biomaven Its quite interesting to think that "if the chinese had it, they would use it offesively" but the americans won't do that... History suggests otherwise.

So this is an interesting example where the AI is clearly better than humans at a very complex task. Need a combination of real smarts and super-human persistence to go through millions of lines of code looking for obscure bugs. If the Chinese state had got this first, they could have penetrated essentially every computer system in the world. So that's why this is a race we can't afford to lose.

@BEBischof I just don't see the point in this kind of posting. If you do not like MCP, no need to develop on top of it, but how does this develop the conversation? What we learn, that everything is so dead again?

Note: I am not starting a band.

@Gijs_Hulsebos @NousResearch Are you focusing on canonical text only or including apocrypha ans surounding texts? Also, how do you solve the issue that original sources are in different languages and also differ from user language, are you using a plan translation?

@NousResearch Building a bible app w hermes rn Truly the word of God ngl

Spread the gospel to your city. Hermes Trismegistus will smile upon you.

@NousResearch I still cannot figure out how can Nous make a business out of Hermes. All profit flies to LLM providers isnt it?

Are you creating a new R package? Think about vignettes as a skill. Most likely your end user will be either a knight (human riding on an agent) or autonomous agent.

@xeophon Barcharts are the piechart of the barcharts class. Cannot compare properly between colour filled groups or show error bars like dodge barcharts.

stacked bar charts are the best chart type, fight me

Many people think that an MCP is just having some tools, while in reality, it is more about model guidance and model education. Think it well, most likely your code is not in the training set, nor is your codebase structure. Design accordingly.

@xeophon Mcp is API + descriptions == CLI + skills. I think we need metrics on token usage of each approach to see if it is really worth it.

MCPs will live by the sheer willpower of companies doing evals about them

@realmcore_ I have been following your work and I think at tjis point you are correct. For end user, the difference at the frontier is not as significant as UX/AX. However, given the very generous allowance of the harnesses AI labs, where do you see the business model of harness makers?

I’m gonna say it again, the models are basically smart enough. It’s the form factor that’s not.

I liked this. Every week: “X is dead, here’s the replacement.” Meanwhile, banks still run COBOL from the 60s or so.

@tangming2005 I think uv solves many pip issues. Besides, now what I do is fresh conda, please claude do as please untill X --version prints.

"Easy to install" is the biggest lie in bioinformatics. I once spent 3 days installing a tool that claimed "just run pip install."

Your LLM is only as good as the data pipeline upstream of it.

@xlr8harder The point is that "this part" becomes a high economic value target and so ai labs and harness companies start targeting until it folds.

Often the disputes seems to boil down to "okay it can do this and this and that but it still can't do THIS part"--but that is just begging the question, no?

Many arguments on what employment impact AI will have seem based in disagreements about what AGI is and whether we will achieve it. The discussion seems much simpler if you start from the premise that there will be a machine that can perform all human intellectual labor.

@Jsevillamol Finally, someone understands me. Slack reply is so awkward like... You have to choose a channel?!

One thing that slack is really missing is a WhatsApp-style "reply" option that quotes the message you are replying to in a thread.