brofallon · HackerTrans

To use RLHF you need a dataset that includes instructions with good & bad answers - do many of those exist? I know there are a few datasets of just plain instructions-with-responses, but I'm not aware of any that have both good and bad (or ranked) responses. Is that trivial, or an important missing element here?

brofallon

3 anni fadiscuss

I agree - LangChain is totally awesome and unlocks amazing new capabilities of LLMs (easily!), but the code itself is pretty awkward. No logging framework, for instance, and the strong typing enforcement with pydantic seems un-pythonic and brittle to me (what if I want to use a different kind of Mappable instead of a dict for something?). Still, great work overall and congrats to the team!

brofallon

3 anni fadiscuss

Somewhat OT, but I feel like this is such an underappreciated aspect of recent LLMs: Not just their ability to generate text - but their apparent effectiveness in making use of arbitrary tools to interact with their environment to achieve some goal. It seems like we're just at the beginning of this with the MRKL and ReAct papers, there will be a ton more awesomeness coming in this area I'm sure.

brofallon

4 anni fadiscuss

There's a special place in my heart for any project in which the Readme says "Warning: The code is fucking awful"

brofallon

4 anni fadiscuss

I'm still a little unclear on the benefits that fusion offers compared to things like wind and solar. I understand that we need to develop better storage technologies for the energy produced by wind and solar, but that seems so much easier than the challenges currently facing fusion. Wind and solar just seem so far ahead of fusion already - they're pretty cheap and very widely deployed on a global scale. In comparison fusion seems very expensive and unproven and even when we get everything to work it might not be much better than a solar farm with a big battery pack. But maybe I'm missing something important about the economics?

brofallon

4 anni fadiscuss

Most of the discussion here seems to assume bioinformatics / genomics jobs are academic, but I work for a clinical testing lab where production-quality code is a must. We're probably a 10/12 on the Joel test.

If you're into bioinformatics or genomics, but aren't excited about an academic setting, take a peek: https://recruiting2.ultipro.com/ARU1000ARUP/JobBoard/62cc791...

We hire fully remote positions and starting salaries are about US$100k.

brofallon

4 anni fadiscuss

Just to clarify, exons are the portions of genes that code for protein sequence (they are expressed). Introns span the distance between exons and may contain regulatory or splicing information. Areas between genes are referred to as intergenic, and also may contain regulatory sequences that affect how genes are expressed.

Part of the issue with centromeric or telomeric sequence is that not only is it hard to sequence (being super repetitive), little is known about what a sequence variant in such an area might mean. It's kind of a chicken-and-egg issue: it doesn't get much attention because no one knows how to interpret variants there, and since no one knows how to interpret variants there it doesn't get much attention

comments