kriro

Listening to audiobooks for me is a learned skill. when I started my mind used to wander off and I couldn't tell you what I just listened to. Now for autopilot activities I'm hyper focused on the books. I have increased the speed to roughly 1.3x which seems to be the sweetspot for my brain. Faster sounds like robot-voices to me and slower made me wander off.

I completely agree with the "mental health" boon to taking time, slowing down and focusing on an activity. This can be done with audiobooks, too. I sometimes sit on my couch and listen to an audiobook, too.

Depends on what you want to read. For fantasy and sci-fi fiction I really like the Youtube community. I did not expect this at all but there are a lot of great "Booktubers" in the 3-5k subscriptions range. Tiktok and Instagram have a lot more "vanity" people and it feels like more of a "look how much I read har-har" kinde of vibe and I avoid them completely.

It is indeed an odd challenge to find good books because I hate to be spoiled and that limits the search options quite a bit.

Fair enough, I understand the UX/interacting with the medium argument. Regarding the processing in the brain, there's claims that "The Representation of Semantic Information Across Human Cerebral Cortex During Listening Versus Reading Is Invariant to Stimulus Modality", see: https://www.jneurosci.org/content/39/39/7722

The study has issues but at least the hypothesis of representing information independent of the modality you consumed it in seems like a decent default to me.

I do not agree with his take on audio books. I find having a book with me and reading at all times as suggested a bit impractical, however for audio books it is way more feasible. I have long-ish commutes of 5h every couple of weeks and listening to audio books on the drive is great. I also listen to audiobooks when I vacuum, clean, grocery shop. Always unabridged versions and I research quite a bit what I want to listen to. I usually "hybrid" listen/read so I'll continue reading the physical book later in the evening.

I keep notes in a physical notebook but keep it super light. I read mostly fiction so it's usually plot points of each chapter and then when I'm done some quick thoughts. I enjoy this and it satisfies my analytical brain to think about things I really liked that an author did or reappearing structures and so on.

I also enjoy the meta-level of reading. Watching review videos on Youtube, thinking about what to read next and finding "hidden gems", so basically the hobby aspect of it.

But a law degree is probably even better. I know what you mean though, consulting companies also hire the (top 1-3%) philosophy majors and math/physics majors for the same reason. Good thought processes.

I find it a bit strange to assume you can only understand these topics with a philosophy degree. My CS degree had a good chunk of philosophy baked in (philosophy of science) and parts of it strongly encouraged you to dive into philosophy. AI 101 introduced me to Gödel for example and logic in general.

From the article it seems like they mostly do "is AI conscious" and ethics work. Call me a skeptic (no pun intended) but it looks like "hiring some philosophers to confirm the things we want to keep saying for the sweet AGI-race-$$$ to flow". Kind of like these tobacco studies way back when.

My anecdotal evidence from France and Germany is that there is very much a trend to look for non-U.S. alternatives which is somewhat recent. It is related to the current government, not really politically motivated but more of a "no guarantees/cannot rely on this partner anymore" thinking coupled with a bit of "WTF annex Denmark, really". Many companies have task forces or projects to investigate moving tech stacks and in the past this was not even an agenda topic (despite strategic advantages and potential cost savings).

It seems a bit odd to me to have a home lab for the sequencing and then feed it to a cloud LLM. I would have thought the point of the home lab is privacy.

And you'd need the DNA of all dogs to boot (unless they are already somehow stored when a dog is registered). But generally speaking, I do like the idea. Also for horse poop which is pretty common on roads here in the countryside.

I loved this game. First RTS I ever played :)

He basically accused many players of cheating without any real evidence. His proofs are notoriously bad. The most notable examples are Daniel Naroditsky who was under immense psychological stress due to the allegations and died due to an accidental kratom overdose (kratom is used to self-medicate anxiety). Another prominent example are his allegations against Jose Martinez (aka Jospem) which resulted in them playing a grudge match which Jospem won. Kramnik was struggling with the tech in the online portion of the game. Basically he cannot accept that younger players beat him in quick formats and flings around ridiculous stuff. He's also feuding with Hikaru who is obviously really good at blitz.

There's a decent argument for using today's words even for settings that are no the current world. But for fantasy specifically I find it breaks immersion quite a bit. I'm not convinced by the "language changes over time" line of argument. If you take it to a bit more of an extreme it would also be fine to have characters say "cool" or even use some gen-z-ism etc, because hey words change meaning over time after all.

However, I'm also aware that I'm kind of a hypocrite because I'm totally fine with current world grammar and punctuation for example.

I guess my pet peeve of "firing an arrow" is also a gun mistake of sorts. It can be found in many fantasy books. A bow is not a gun, there's no gunpowder involved. If a commander orders his archers to "fire" what should they do, set their bows on fire?

Is there a list or overview of all open source projects that refuse to accept AI-code? It pops up every now and then but would be nice to have a good reference. Could also be interesting for quality analysis and so on in the future.

I can fully understand the reasoning for it and I also think it's by and large a good idea because one of the cool things about open source projects is that they are traditionally good places for younger developers to get started or hop in and learn. There's a lot more to be learned from writing everything yourself and thinking through it.

Yeah just to clarify, I meant there are probably a lot of issues in existing smart contracts that can be discovered with Mythos style LLMs. And given how many exist and are fairly blindly trusted (who ever reads the audit reports) this could be a pretty epic meltdown spiral that gets triggers.

The fun ahead will probably be capable LLMs + smart contracts. There are probably a lot of issues that can be discovered.

I wear over-ear headphones almost all the time when I'm out on my own using public transportation, grocery shopping etc.

I listen to audio books (fiction) and it's great. I'm an introvert and this actually helps me keep high energy levels all day long. Another plus is I'm usually immersed in the story and not using my phone (well apart from it being a playback device) when sitting down or waiting somewhere.

And it's precisely this extra energy I can use to have more meaningful interactions with other human beings. I don't wear headphones when I go climbing for example and interact with random strangers quite a bit when I do.

I also don't appreciate the stereotypes that are flung about in the article. I'm also German, plenty of interactions as described in his Jalapeno-Story all day every day.

That jump through the mirror blew my mind as a kid. And level 6 oh man took us forever to beat that slightly obese guard. I still remember that we once accidentally walked all the way into that guard and switched places but then were killed by a strike to the back. always wanted to replay and see if I can pass the guard that way.

There's a certain irony in coining the term on discord. Nice blog post but I'm used to reading these from people who hang out on IRC. Times are changing indeed.

My private version of anti-dopamine fracking is playing the phone game. Every social event I attend, I try to be the last person to look at their phone (well basically not look at it at all). It is fairly sad how easily this game is won in under 30 minutes in most casual settings.

I've been a bit out of the loop with Austrian Economics (last re-read of Human Action was ~15 years ago). I'm very well read in it and enjoy the aesthetics of the theories and the history of thought books but got very tired of the online flame-wars and the political side in general (both the pro- and anti-Austrians). So Praxeology of Privacy sounds like an interesting read, I'll give it a go this year.

Unfortunately DDG is still horrible for non-English results. As are most "smaller" search engines. I rotate through them every now and then to try. Is there a meta search engine that uses country specific engines depending on searches anyone can recommend?

We're talking about enterprise customers. The trivial answer is Mistral has sales teams and consultants from the same company that builds the models and from the EU.

I don't see it mentioned explicitly in the methods section but I assume you prompted each model only once for each question? Did you consider prompting n-times in blank states to see if the models even agree with themselves?

Would also be interesting to add a virtual model that is simply the majority of all models and see how much the individual models differ from the "consensus".

Do you plan to add some sources in the related work section of baseline numbers for human expert disagreement in fact checking tasks (I'm assuming such studies exist).

I agree. Once upon a time I was quite interested in FPGAs but the infrastructure being so uninviting in general made me move on completely. I was somewhat recently involved in quantizing neural networks with FINN (AMD) and let's just say...that was a pretty bad experience overall.

Nice analysis, I would have loved a short overview of the kinds of experiments that were running on the machine (I know the results are given).

I find the "independent researcher" business model quite interesting. In the linked post he writes """DFT is a proprietary training algorithm, however, I’m currently offering a beta for a model training service where I will train your model for you using DFT.""" I'm curious how successful this is. Essentially market some AI breakthrough as a service instead of publishing a paper like my academic brain is trained to do.

As an aside, one thing that I always loved about our field was that the startup cost for many business ideas was "a laptop, internet connection and some some grit". In the age of AI it's quite a bit more and I feel one of the sad side effects of this is that it crowds out poorer and younger developers.

If I'm only thinking about non-programming business applications, anecdotally, Mistral is certainly a player in the European enterprise market. For most German companies I have interacted with, Mistral was the first point of contact regarding corporate AI rollout. For the "small potatoes" day to day minutia Copilot is probably the #1.

I used it a bit, had it installed for a while on a G4 PowerBook (must have been early-ish 2000s). I like the no-nonsense attitude towards blobs, security focus. Overall the experience was very good. The bit of code I read was also written nicely. I'll always endorse it and should really install it somewhere again in the near future.

This is also the 60th release. Congrats team.

Seems like textbook Inside the Tornado marketing. Pick a country as a bowling pin, show some success, go for a different/bigger country. Presumably cover EU first this way. Be the first to offer all-citizens licenses.

I did it back in the day when fast.ai was relatively new with ULMFiT. This must have been when Bert was sota. The architecture allows you to train a base and specialize with a head. I used the entire Wikipedia for the base and then some GBs of tweets I had collected through the firehouse. I had access to a lab with 20 game dev computers. Must have been roughly GTX 2080s. One training cycle took about half a day for the tokenized Wikipedia so I hyper parameter tuned by running one different setting on each computer and then moving on with the winner as the starting point for the next day. It was always fun to come to work the next morning and check the results.

The engineering was horrible and very ad-hoc but I learned a lot. Results were ok-ish (I classified tweets) but it gave me a good perspective on the sheer GPU power (and engineering challenges) one would need to do this seriously. I didn't fully grasp the potential of generating output but spent quite some time chuckling at generated tweets (was just curious to try it).

Innocent mistakes and frustrating back and forth are also very common, especially for interdisciplinary teams. The mismatch in tooling and workflows and manual copy & paste conversion is a thing to behold. Add multiple countries and Excel to the mix (dot vs comma, formulas being language specific), maybe have a couple of Chinese, Japanese, Russian or Arabic speking researchers in the group for some extra UTF-8 magic. Line endings on Linux vs. OSX vs. Windows.