pugio

I've read most of Stephenson's books, and loved many of them, but the only one I absolutely tell people to never read is the Rise and Fall of D.O.D.O. (despite the signed copy on my shelf). I haven't read it since publication, so the details are a bit hazy now, but I remember it feeling like a waste. The conceit is: What if there were a government organization that dealt with <cool scifi thing>, and we emulated that with high fidelity – complete with lots of reporting and bureaucracy . To me, it made <cool scifi thing> about as fun to read about as... tedious government bureaucracy.

It's nice to hear that other people really liked it. Definitely highlights the breadth of approaches and styles that Stephenson has.

It could very quickly become racist, but it isn't necessarily. The Masters of Scale podcast recently did a nice episode on using a voice clone of Reid Hoffman to help with various dubs and edits: https://mastersofscale.com/meet-reid-ish-the-story-of-clonin...

There's a segment where the various staff/producers all try to imitate Reid's style, and then apply the voice transfer model to here their impression in his voice.

I've also used an open source TTS model conditioned on an actor whose voice I like to read the transcript from a podcast host whose voice I have a hard time tolerating. Here's where you start to edge closer to possible racism, echo chambering, etc. It's certainly a gray area, and I haven't navigated the moral shoals and reefs of replacing one voice with another* I like more – which is why I won't publish anything about how I did it.

(*A random sample: Better to not listen to what a person has to say, or to listen to your voice-cloned version of their words? What about if you literally can't understand them? Non-native speakers already rewrite their words using ChatGPT. Possibly (likely) only okay if they do it themselves...)

It sounds like you're viewing "OOM" as "Out of Memory", when the parent meant "Order of Magnitude" - meaning that a baby's "training data" is several Orders of Magnitude more than what the DenseAV model from the paper used.

As someone who periodically listens to papers via text-to-speech, I was intrigued when I saw this link. I'm a little uncertain about the "AI simplification" - If I'm reading a paper I don't want to have the second guess everything I hear as to whether it completely misinterpreted, elided, or hallucinated a point.

Do you have any personal experience with the results?

This is the perfect follow-on to your joke: https://www.youtube.com/watch?v=vv445S62dgU ("When People Hear You Have A Hobby").

I would really love an `ocrmypdf` like tool which uses Apple Vision to create searchable PDFs from scanned images. I've been searching every week or so for some kind of project but so far haven't found anything. Perhaps it's time to make it myself...

I've spent the last couple weeks diving into various PDF parsing solutions for scientific documents. GROBID is pretty cool, but it made some mistakes when trying to parse (I think arxiv) papers which removed some of the text.

Even though it gave a lot of great structured options, missing even a single sentence was unforgivable to me. I went with Nougat instead, for arxiv papers.

(Also check out Marker (mentioned on hn in the last month) for pretty high fidelity paper conversion to markdown. Does reasonable job with equations too.)

> We’re currently using Mixtral 8x7B behind our endpoint mistral-small...

So 45 billion parameters is what they consider their "small" model? I'm excited to see what/if their larger models will be.

That's me exactly. I used to eat half a bar of 85% - 100% dark daily. The (relatively) recent info on heavy metals has me really bummed. Why is everything always so toxic these days???

My new goal is to figure out how to get all this heavy metal accumulation out of me (I've been eating heavy dark chocolate regularly for years), and find ways I can keep eating chocolate without absorbing more. Some studies I've read mention that tomatoes might help prevent heavy metal absorption when consumed with the source.

How come the blog post doesn't discuss training at all? Which data sets were used? How many tokens was it trained on? I couldn't seem to find a linked preprint or anything else really

This resonated with me, thank you for writing it. My own personal rule now is that video game playing should also be social time. I started playing BG3 using wine (on Mac) when it came out, but only with friends. Gaming time is (crucial) social time.

Eagerly anticipating this game has brought me back to my childhood days when playing the game I wanted seems like the ultimate fun, but of course now I have responsibilities. Now playing is a reward for myself if I handle everything else and again, only with friends. Even though I've been tempted dozens of times in the last month to fire it up for a solo session, that rule has stood me in good stead.

I'm hoping that the Apple Game Porting Toolkit signals a shift to more games support.

Thankfully, the Apple Silicon machines are powerful enough to support many modern games. I'm currently (and happily) playing Baldur's Gate 3 using Wineskin on my M1 MBP with no problems. So far no bugs, and multiplayer with PC users works great.

I don't normally engage with these kinds of comments but today I guess I'm extra curious: why is "paid sick leave" a "basic human right."?

I don't disagree that it's a good and moral thing to offer, I'm just curious about where these universally agreed upon "basic human rights" come from?

I think you may have misunderstood the OP and actually agree with him. He's saying that AAA studios (sometimes) spend more money to produce lower quality products, and that claims from some AAA studio devs that you shouldn't use BG3 as a new standard are just them trying to excuse their own mediocre offerings.

Truisms aside, in many cases the absence of a thing which one would expect to see is in fact a support for alternate hypotheses.

ie. "Given a world X is true, I would expect to see Y. Since I don't see Y, that's at least weak evidence that X may not be true."

There are of course plenty of other reasons why you wouldn't see Y, but it's at least some kind of signal.

I've found that Google's reporting severely under reports air quality (making it look better than it is). The purple air map itself seems much more reliable for tracking actual local air quality.

I use purple air zoomed out combined with windy.com to generate my own smoke predictions, it's worked pretty well for me.

Every 3 hours. But also, if you have API access you can use the OpenAI playground to query GPT-4 as much as you want. Sure, you pay a cost (per token), but if you're willing to front the bill you can move pretty quickly.

Looks very cool. I am also working heavily in this space (using AI to help learning).

One minor point: In your demo, you state that "Today, almost 50% of code is written by AI, so why shouldn't you learn how to code with AI?"

Citation? I thought this might be true of developers using Copilot (based on their recent publications), but even that seems a bit of a stretch. I guess you could claim this because it's so easy to generate LOTS of code with an AI model. How much of that code is used, though? Really 50% of ALL code written now?

I have it running, slowly, on the same machine. I would love for someone to get support running for MPS backend (The GPU) but it does run on the CPU.

My initial impression is (compared to ChatGPT 3.5-turbo) - not impressed.

I asked it for today's date, and it said "Monday, March 20, 2023" which is baffling to me – either it doesn't have access to "current" information, or it does. But off by one?

"What is the knowledge cutoff of your training data?" > My training data was cut off in September 2021. This means that I do not have access to any information that was published after that date. However, I am able to access and process information that was published before that date.

Then I tried to get it to help me with finding some papers, but it continuously misinterpreted my prompts, or gave me lackluster results.

Back to GPT I go.

I thought this was a nice analysis of the study: https://peterattiamd.com/more-hype-than-substance-erythritol.... Since our bodies produce erythritol as part of the pentose phosphate pathway (which is activated heavily by many of the same factors which increase the risk of heart disease), it's possible that high circulating levels in our blood are just a marker/signal of other factors.

Thanks for the links, this is pretty interesting stuff. Just a quick note: it's Conflict Free Replicated Data Types - not Relational.

Seems like it's only for Windows... except all the screenshots clearly show Macbooks / iMacs.

I like the idea of the service, but wouldn't use it unless I can run it on Mac. Streaming for indie games would be perfection, as I don't have the time to go deep into a AAA title but would love to quickly start up some cool short indie game on demand.

I don't really use Twitter and also prefer blogs to reading tweet threads, but one interesting advantage of the Twitter model is that it allows easy quoting/replying/liking of individual paragraphs / sentences.

I'm not sure if this outweighs the different reading experience, but that ability makes for something which is fundamentally different, rather than being a strictly worse version of a blog.

I've been thinking (and researching) heavily in preparation for a new (offline, desktop) app I want to build. The app has a lot of data wrangling, and probably a decent amount of ML. Python seems like the logical choice, but...

I've written a lot of Python, taught classes in it, deployed production code... and I still feel a semi-conscious urge to reach for something else whenever I contemplate starting a new project. Something about its approach, syntax, common idioms, always feels just a tad clunkier to me than I'd prefer. The whitespace makes it hard to paste lines into my repl, and (probably the biggest thing) comprehensions ARE NOT SUFFICIENT replacements for easy anonymous inline functions (JS () => , or Ruby's blocks).

I like dynamic languages, I really like the advances in tooling with type checking and LSP support. I like dynamic notebooks (either inline in VS Code #%%, or straight Jupyter), the massive package ecosystem (but obviously hate the actual packaging tools), cool tools like Rich... but it just doesn't make me as happy to use as other languages.

I'm still trying to articulate why. Maybe I was just ruined by learning Ruby first.

I've read and loved most of the fantasy you mentioned, except for saga of The forgotten warrior. This makes me an excellent target for that last recommendation, thank you!

It's worth noting that the narrator - Michael Kramer - is exceptional. One of the characters in this particular series is known for his ability with accents, so listening to the audiobook with a good narrator is an extra amount of fun.

I've read most of Sanderson's books in print, but for this one I also bought the audiobook specifically for that kind of narration.

My family and I have been supremely careful Mar 2020. I read a lot of the research and preprints and am concerned about long term system impacts of repeat infections. Despite huge numbers of precautions, we all finally caught it about ten days ago. Still positive, still coughing, still tired.

We can lower risk, but it seems that no level of caution is truly enough, given that others don't care. I'm really not sure what to do – we don't want to keep getting it, but we really want some sense of social life back. My wife and I keep wondering / looking if there's anywhere in the world where people by and large take this infection seriously. I suspect it will come down to small communities, rather than states or countries.

I was a little surprised the article didn't mention declining sperm counts and testosterone as a possible factor.

My impression was that increased exposure to toxic compounds (microplastics, estrogen mimicking compounds, etc) has been contributing to declining sperm counts / motility, and things like earlier onset of puberty in women.

Wouldn't those at least have some effect on sex drive? I'm looking in particular at the decrease in adolescent masturbation, as it's hard for me to imagine that "increased video gaming" is a culprit for this.

For those looking to start understanding the fundamentals of synths and sound, check out https://learningsynths.ableton.com/en/get-started (and the recently posted https://ciechanow.ski/sound/ ). A great little explorable on the basics.