Comments URL: https://news.ycombinator.com/item?id=46722853

Points: 147

# Comments: 76

Comments URL: https://news.ycombinator.com/item?id=45118963

Points: 1

# Comments: 0

Comes down to: - Self selection bias - Trial design - Dubious intepretations of neural connectivity

Their trial design and interpretation of results is not properly done (i.e. they are making unfair comparison of LLM users to non-LLM users), so they can't really make the kind of claims they are making.

This would not stand up to peer review in it's current form.

I'm also saying this as someone who generally does believe these declines exist, but this is not the evidence it claims to be.

Also back when I had to do these (I used Wave) having a notes section was very useful to include a few things (i.e. I used to include conversion rates). Would probably be pretty easy.

You can make a strong argument these institutions require reform, but such reform should be done not overnight, and not through such broad strokes.

2. What is illegal about the change. The NIH overhead rate is actually negotiated directly between the institution and the NIH, following a process put into law. This is why a federal judge has blocked this order [1]. I'm far from a lawyer, but my read of this is that this is a change that would need to come through congress or a re-negotiation of the rates through the mandated process.

[1]: https://www.aamc.org/news/press-releases/aamc-lawsuit-result...

For those who are questioning the validity of a 59% (or higher for some other institutions) overhead rate, your concerns are worth hearing and a review could be necessary, but oh my please not like this. This was an overnight (likely illegal!) change made with no warning and no consultation.

If the government decided that a cap was necessary it should be phased in to allow for insitutions to adjust the operational budgets gradually rather than this shock therapy that destroys lives and WASTES research money (as labs are potentially unable to staff their ongoing projects). A phased in approach would have nearly the same long-term budget implications.

Are there too many admin staff? Likely? Is this the right way to address that? Absolutely not.

For those who are unfamiliar with how career progress works in Academia, it is so competitive that even a year or two "break" in your career likely means you are forever unable to get a job. If you're on year 12 of an academic career, attempting to get your first job after your second (probably underpaid) postdoc and suddenly there are no jobs, you can't just wait it out. You are probably just done, and out of the market forever as you will lose your connections and have a gap in your CV which in this market is enough to disqualify you.

TL;DR is three major factors:

1. The agencies that are doing the estimates are _very_ bad at exponential development curves (cough cough IEA estimating solar [2])

2. Unfortunately much of the developing world's economy is not growing as fast as we previously thought it would (similar thing happening with birthrates)

3. Many costs are absolute and _not_ marginal, which is just wrong IMO. We are going to need the energy either way, we should be talking about the "green premium" (as far as it exists), not how much it'll cost to generate XX TWH of energy

[1]: https://www.economist.com/interactive/briefing/2024/11/14/th...

[2]: https://www.economist.com/interactive/briefing/2024/11/14/th...

This is, for better or worse, the kind of research you can only do at institutions which have free fMRI scanning (MIT, Princeton, Harvard, etc.). No behavioural links, only very detailed activation maps that we can't really draw conclusions from. A missed opportunity IMO to spend these kinds of scanning resource on some kind of more narrowly focussed task with some behavioural outcome they can link to brain data.

Most of SWE (and much of engineering in general) is built on abstractions -- I use a Numpy to do math for me, React to build a UI, or Moment to do date operations. All of these libraries offer abstractions that give me high leverage on a problem in a reliable way.

The issue with the current state of AI tools for code generation is that they don't offer a reliable abstraction, instead the abstraction is the prompt/context, and the reliability can vary quite a bit.

I would feel like one hand it tied behind my back without LLM tools (I use both co-pilot and Gemini daily), however the amount of code I allow these tools to write _for_ me is quite limited. I use these tools to automate small snippets (co-pilot) or help me ideate (Gemini). I wouldn't trust them to write more than a contained function as I don't trust that it'll do what I intend.

So while I think these tools are amazing for increasing productivity, I'm still skeptical of using them at scale to write reliable software, and I'm not sure if the path we are on with them is the right one to get there.

Comments URL: https://news.ycombinator.com/item?id=41680853

Points: 1

# Comments: 0

She has found that prolific is substantially better (you have to pay more for it as well), however that may only be because it's a higher cost/newer platform.

However, after a few months the ROI crept down until it wasn't worth it anymore (access patterns stabilized). I'm tempted to bring it back once and a while but the price tag keeps me from having it always on.

What is interesting is that despite these constant misses, the IEA is constantly used as the definitive source for energy modeling across the world.

Comments URL: https://news.ycombinator.com/item?id=36712580

Points: 2

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=35766342

Points: 1

# Comments: 0

As an aside, what is the issue with versioning these days? Ruff and FastAPI both have massive user bases and a reliable codebase, but haven't released v1.0.0! Ruff hasn't even got as far as v0.1.0.

[0]: https://techcrunch.com/2023/02/16/sequoia-backs-open-source-...