samvher

I've been testing these models with fireworks.ai

They provide serverless access, and if you use fireconnect it's also easy to hook the models up to e.g. codex which allows for fairly clean comparisons

I think they provide the bare model (no filter) but I might be wrong, haven't tried using it for security research yet

Have you been able to find information on which languages are explicitly supported by this API? I'm having a hard time finding that out.

Yes and many forms of electricity generation evaporate water as well

I also noticed this on 2 accounts - did not take as careful notes as you did unfortunately. But I'm fairly confident - both of these are accounts where I care about the interactions not being used for training.

What's kind of still an open question for me is if the toggle automatically also applies to my Codex CLI use on the same account, or if that data is still silently being used in some way.

After this happened, I deleted all my ChatGPT history (even though I'm not sure how much it helps at this point), but for Codex I still haven't really found any way to do the same, I can still load my past sessions even after archiving them.

Interesting challenge.

1 - I worked abroad and wasn't really familiar with the systems there. Gemini made me aware of a kind of pension account that I could withdraw from when I left the country netting me a few thousand dollars.

2 - Working as a tech contractor, charging by deliverable, Codex/Claude Code speed me up and it doesn't seem to have significantly dropped rates in the market.

3 - Also contractor related: I had Claude do a quick legal sanity check of my contracts, and it warned me of some clauses that I'd be better off removing/changing/refining. I was not aware of these nuances and would not have paid a lawyer for this as the contract was too small, but the changes were accepted by the client and reduced my risk exposure meaningfully.

4 - Learning a foreign language, I use it to check my draft emails and messages. It corrects them but also serves a tutoring role providing feedback, improving both the accuracy of my communication and my rate of language acquisition.

5 - Gemini Deep Research helped me narrow down tent models that met my fairly specific set of requirements. Very happy with the tent I ended up buying, from a brand that was not on my radar before.

Yes I was very surprised after the whole "scandal" around ChatGPT becoming too sycophantic that there was this massive change in tone from the last preview model (05-06) to the 06-05/GA model. The tone is really off-putting, I really liked how the preview versions felt like intelligent conversation partners and recognize what you're saying about useful pushback - it was my favorite set of models (the few preview iterations before this one) and I'm sad to see them disappearing.

Many people on the Google AI Developer forums have also noted either bugs or just performance regression in the final model.

https://ppc.cs.aalto.fi/ covers some of this (overlapping with the topics the person you responded to mentioned, but not covering all, and including some others)

If the only moat really will be the scale of computation resources, that's great news for users, because it will be an extremely competitive market where prices will be driven down very effectively.

I suspect that model quality/vibes and integrations will play a role as well though.

How did you go about learning this? It's something that I've been thinking I might be interested in picking up.

I can't tell if you're insinuating that Singapore is a pass-through for H100's heading towards China or whether there is some significant development taking place in Singapore that I'm unaware of?

What's happening definitely makes me nervous, but "at best a WW3-event and at worst an extinction-event" seems a bit much. Mainly because there are a _lot_ of unknowns. Better try to get comfortable with just riding this out.

I'm trying to parse the idea of "a collective agreement" but can't fully wrap my head around how that would work.

It seems to me more like the lack of a "Walmart Law" is a result of e.g. lack of economies of scale and other economic structure, rather than some collective agreement. (If it was profitable to break out of that agreement and start a "Walmart Law", it seems we'd see that happen pretty quickly?)

But if you know more about this and I'm off the mark I'd love to learn

For me the excitement is that around the o3 announcement I had a feeling like we were heading to an OpenAI / Sam Altman controlled dystopia. This resets that - you can run the model yourself, you can modify it yourself, it's essentially on par with the best public models, and it gives hope that the smaller players have a fighting chance going forward. They also published their innovations bringing back some of the feeling of open science that used to be in ML research but which mostly went away.

Given what we just saw in terms of the DeepSeek team squeezing a lot of extra performance out of more efficient implementation on GPU, and the model still being optimized for GPU rather than CPU - is it unreasonable to think that in the $6k setup described, some performance might still be left on the table that could be squeezed out with some better optimization for these particular CPUs?

I don't think the part about the market reaction is too surprising, for COVID it took a while for the market to react as well.

Seeing that DeepSeek-R1 has benchmark scores like o1 is not the same as seeing that people actually like it, that it's being adopted, and that stated training costs are getting accepted as credible.

(But I agree that there is a lot of irrationality in these markets.)

It crashed all the way back to June 2024 levels, eons of progress wiped out

My interpretation is that yes in the long haul, lower energy/hardware requirements might increase demand rather than decrease it. But right now, DeepSeek has demonstrated that the current bottleneck to progress is _not_ compute, which decreases the near term pressure on buying GPUs at any cost, which decreases NVIDIA's stock price.

A bit like humanoid robotics - not the most efficient, cheapest, easiest etc, but highly compatible with existing environments designed for humans and hence can be integrated very generically

What kind of preparation are you suggesting?

I followed the drama. The point I was (somewhat unsuccessfully) trying to make was that while, sure, there were groups who wanted him back (mainly the groups with vested financial interests and associated leverage), my sense was that the way it played out was not necessarily in line with wider humanity's best interest, i.e. as would have been hoped based on OpenAI's publicly stated goals.

Define everyone. I was delighted when they fired him. I don't believe he has humanity's best interest at heart.

The first few years out of university I consistently had < $10k on my bank account and was traveling as much as I could, with just occasional part time work to "fill up" a few k$. I loved it and couldn't imagine living any other way, neither retirement nor risks were just really on my mind at all (it wasn't really a zen thing).

Now I have a child and feel completely different. The moment she arrived I immediately felt way behind on retirement savings and stability and since then my #1 priority has been to catch up. I'm hoping that not long from now I'll be able to translate being more secure on paper to also mentally getting back the care free feeling I had before.

What does issues with relationships mean in this context?

Also, it sounds like they were traveling for 5 years? Yeah, that does sound like a very long time - I imagine if it had been 1 year or so it might have been a very different story (?)

100 years later, sure. What about in December 1924?

Is ratadj a misspelling of something? I've spent some time in EA circles and I don't know what it is, wondering if I'm missing something.

Seems like there might be some survivorship bias here, right? You teach students who made it to a top university because they thrived in the classical setup which is the most common one. Presumably your preferred teaching style aligning well with the classical approach also helped get you to that position.

Personally I feel like my education/learning only really started to take off when resources like EdX and Coursera became available. I did reasonably well at uni but was not motivated with others deciding what I had to learn and when. Lectures tended to be slow paced and often boring, so I zoned out instead of being pulled in (I passed my exams by working through the problem sets in the textbooks, I skipped most lectures).

When I got the ability to play/pause/skip/1.5-2x videos, and when I could choose what subject to learn like a kid in a candy shop, I did start consuming lectures much more aggressively. Still, I think well designed problem sets and assignments actually do the bulk of the work when it comes to learning/teaching, and I regularly skip the lectures and dive right to those.

Not saying that your method doesn't work, or that it shouldn't work for you, but its suitability depends on the topic, the student, and the setting.

I am interested, would be happy to connect in whatever way works. My email is on my profile.

> I always come back to AlphaFold [an AI system from Google DeepMind that computationally predicts protein structures]. In 75 years we figured out 10,000 protein structures; AlphaFold has figured out 200 million. If it were human, it would have won a Nobel Prize.

Haha!