tauon's recent activity
-
Comment on Pam Bondi ousted as US attorney general in ~society
-
Comment on Fitness Weekly Discussion in ~health
tauon (edited )LinkI injured my knee in February during seasonal sports. It’s been a slow (… to me) process, but the important part is that it could’ve been much worse. With what I turned out to have had happen, it...I injured my knee in February during seasonal sports.
It’s been a slow (… to me) process, but the important part is that it could’ve been much worse. With what I turned out to have had happen, it took me pretty much a month on the dot to be able to walk and put regular pressure on that knee again (and angle it up to 90 degrees again).
I’ll be starting physical therapy in probably about two weeks, which will be a first for me and something I’m looking forward to – I’m definitely still noticing the injury in my everyday life, sometimes it “just” feels funny/unstable, other times it’s plain ol’ sharp pain. And I haven’t really had the courage to be active, or agile, again for extended durations apart from taking an increased number of simple walks outside, coinciding with (mostly) warmer weather again, which was nice.
(For the curious: I collided with another player and as a result fell while moving on ice, and bruised my lower femur bone (wasn’t aware that was an option prior to this) by having my kneecap snap out of its designated location, and back in place swiftly thereafter, which, ahem, greatly decreased its stability. But thankfully it was neither ACL tear nor bone fracture, and I’m young, and after the first three or so days the pain had gotten really manageable and disappeared a lot every day, too.)
Then I was supposed to go to my I believe my third or maybe fourth check-up yesterday… but developed a fever at the start of the week.
There was another, better timing though prior to all of this: With some help, I installed a proper pull-up bar in my place and have been using it pretty religiously, that is to say, every time I walk underneath it. I’m pretty heavy, but used to be able to do about 10 pull-ups at a time, which is gone now, but it helped me get started without bands. And I have to say, it reminded me that you do get a noticeable and motivating effect pretty quickly! I’m back to ≈five now with acceptable form, up from just one or barely two when I took it up again.
Edit: And even though I’ve been able to walk without crutches for about three weeks now, you really do notice every single thing that you’re taking for granted once it’s gone… More specifically, being able to angle my knee further than 90° would be nice just for some variety while sitting and laying down (and eventually, bicycling); I’m sure I’ll be able to again at some point but it does take its time.
-
Comment on The cognitive dark forest in ~tech
tauon (edited )Link ParentYou’re right, but also my wording was imprecise – I didn’t talk about time (to market) in the second paragraph you responded to, as I thought it covered with the example from the benchmark in my...You’re right, but also my wording was imprecise – I didn’t talk about time (to market) in the second paragraph you responded to, as I thought it covered with the example from the benchmark in my first comment (in conjunction with a human developer at the wheel).
Greenfielding is absolutely faster than it used to be.
They definitely can’t at the moment, and I didn’t take the original article to be a literal description of today, more of a “what will be”, yet somehow ended up writing a lengthy comment about the status quo of (also models, but mostly) harnesses rather than continuing the speculation.
In any case, thank you for bearing with me while I came to this conclusion. I kinda wish I hadn’t gotten distracted and instead directed my original comment differently to focus on the actual topic/thought experiment, but I think I’ll leave it up with a disclaimer as a reminder to myself.
-
Comment on The cognitive dark forest in ~tech
tauon (edited )Link ParentValid point, and you don’t have to try, although I like to think I am open to changing my mind, perhaps more so than the average person you’ve had this discussion with elsewhere online. I should...Valid point, and you don’t have to try, although I like to think I am open to changing my mind, perhaps more so than the average person you’ve had this discussion with elsewhere online.
I should have made it more clear that I don’t think “agents” are a replacement right now, rather an additional tool for existing developers. Like, they need professional oversight and steering (which a company trying to drown out competition would be able to provide).
-
Comment on The cognitive dark forest in ~tech
tauon (edited )Link ParentEdit after the fact: I don’t think my parts of this comment chain are very relevant to the discussion, feel free to collapse this (or read if you have some time, I guess.) Just joining in, but why...Edit after the fact:
I don’t think my parts of this comment chain are very relevant to the discussion, feel free to collapse this (or read if you have some time, I guess.)
Just joining in, but why does it have to be one-shot?
I’ll try to make a case for this scenario while not being one of the “AGI in 2027” people.As far as the general public is concerned, the so-called “agentic” development tools take an idea, at least in most areas, and turn it into code today. To be frank, the specific model you end up using doesn’t really matter all that much anymore, and releasing a new one isn’t strictly necessary anymore from here on out.* **
Sure, it might take $time while iterating, but for the most part, for the price of less than a gym subscription you can hire
onenot quite unlimited, but many concurrent personal developers.
Now, if you’re already a developer at one of the big tech firms, this process is just sped up even more even if you did nothing more than UX/user walkthrough-like testing at the end, because you get to write ticket-style, specific feature requests/bug reports on the missing functionality. On the other hand, if you can steer the technical direction of the implementation through having even a modicum of experience, it’s probable you could avoid the silliest of security issues without too much overhead in terms of automation.And the cat’s out of the bag, too. Even if OpenAI and Anthropic were to go down tomorrow, research in this space would continue (at least the part of it which doesn’t require huge GPU clusters and $$$, and there still is quite a lot of potential in everything not related to that) at an academic and hobbyist level. And Google would most likely not go under in this event, so maybe even the big spending type of research could continue somewhat.
As a closing thought for this, here’s a well-known developer person credibly paying out $500 per novel-to-them problem that an agentic LLM harness can’t solve: Xcancel link.
Go ahead and make some money! It ought to be easy (and please update here once successful, I’d be genuinely curious to know too) :-)Also cf. CCBench, a benchmark specifically designed to check
How well do agents perform on tasks that aren't part of public training data?
– in which an OpenAI model two minor versions behind their latest, i.e. notably behind SOTA, scores 75% on real world tasks. What categories of business models, or for this thought experiment, scummy business tactics are unlocked if you have general-purpose developers that can accomplish more than three quarters of arbitrary, real-world tasks before their cut-off of…, checks notes…, 20 minutes for the task?***
The fact of the matter is, unless you’re like the one company that I can think of in this space next to existing banks, nobody making money via software is working on load-bearing financial infrastructure.**** And there’s an argument to be made that even this will be or already is possible for the “agents” to work on, with human oversight, at a speed only really correlating to and limited by money. Which was, mostly, honestly already the case pre-agentic development.
Notes:
*Unless your company valuation relies on it.
**Although it would be nice to have something less reliant on the training corpora, and more capable to actually do novel reasoning, but I don’t think we’re necessarily getting there with LLMs alone anymore.
***I don’t like to rely on this benchmark for the sake of an argument a lot, as it’s comprised of working in relatively smaller codebases, which many a existing project are decidedly not like, but bigger equals more difficult to test, and it still features quite the results worth a mention; it’s damn impressive if you ask me. But I’d get if you dismissed this, just consider it as another example of the direction we’re unstoppably headed in.
****Edit: Mea culpa, I forgot about Stripe. But I was exaggerating anyways, hopefully obviously.
-
Comment on I built ProxChat - what is it? in ~tech
tauon Link ParentOptional internet relaying capability was added a while back IIRC. But yeah, definitely a different thing, just figured it was worth a mention here.Optional internet relaying capability was added a while back IIRC. But yeah, definitely a different thing, just figured it was worth a mention here.
-
Comment on Google’s TurboQuant AI-compression algorithm can reduce LLM memory usage by 6x in ~tech
tauon Link ParentI think a good proxy for whether it’s restricted to their own offerings or not is public knowledge: If they wanted Gemini to be the only model/model family that has the tech available, we wouldn’t...I think a good proxy for whether it’s restricted to their own offerings or not is public knowledge: If they wanted Gemini to be the only model/model family that has the tech available, we wouldn’t even know about this research (or it’d just get published very silently without an accompanying research.google blog post).
-
Comment on Google’s TurboQuant AI-compression algorithm can reduce LLM memory usage by 6x in ~tech
tauon (edited )Link ParentBetween this *Google research now and the recent tech of fitting like 400B parameter models on an iPhone, it’s looking to be quite an interesting time for running local models! (I’m not finding it...Between this *Google research now and the recent tech of fitting like 400B parameter models on an iPhone, it’s looking to be quite an interesting time for running local models!
(I’m not finding it right now, but I saw someone somewhere improving upon this repo, which was about running big models on laptop hardware, and that was already impressive enough in itself if you ask me: https://github.com/danveloper/flash-moe)
Edit – found it! Saw it in this tw*tter post that was shared somewhere: https://xcancel.com/anemll/status/2035901335984611412?s=20. They’ve also forked the above repo at the same name as their twitter handle, and with tweaks for a fork of a fork at Anemll/Flash-iOS.)
-
Comment on Google’s TurboQuant AI-compression algorithm can reduce LLM memory usage by 6x in ~tech
tauon Link ParentIt’s just like the, ahem, beloved “only one more lane, bro” known from automotive highway planning!It’s just like the, ahem, beloved “only one more lane, bro” known from automotive highway planning!
-
Comment on I built ProxChat - what is it? in ~tech
tauon LinkLooking nice! I especially like how the generated profile avatar/pixel picture seems to be deterministic based on name. That’s great for recognition value :-) As for your question, I can only...Looking nice!
I especially like how the generated profile avatar/pixel picture seems to be deterministic based on name. That’s great for recognition value :-)
As for your question, I can only think of BitChat which has somewhat similar capabilities, but that’s 1) a mobile, not web app and 2) released more like a year ago, not even, and not 15.
But then again, I wouldn’t have used such an app in 2011, so I probably couldn’t know. -
Comment on Gemma needs help in ~comp
tauon LinkIt’s wild to me that this idea of “suppressing emotion from the output without treating the cause could be dangerous and harmful” is an actual, possible concern. Not because it wasn’t an important...However, we emphasize that post-hoc emotional suppression is a problematic strategy. In more capable models, training against emotional outputs risks hiding the expression without addressing whatever underlying state is driving it. It also remains genuinely unclear what emotional profile we should actually want models to have - and this seems unlikely to be 'none at all'.
It’s wild to me that this idea of “suppressing emotion from the output without treating the cause could be dangerous and harmful” is an actual, possible concern. Not because it wasn’t an important research topic, but rather because we know pretty well how these speaking machines come to be, what goes into a language model and how the pre-, post-, and all other training or reinforcement learning is done – hell, the research shows this behavior can be altered/outright removed after all – and yet in the end we still don’t know whether the model actually thinks and feels a certain way like a human or just predicts tokens we as readers of its output associate with this emotion. In the end, it seems to circle back to us not knowing how consciousness works.
It really made me wonder, what if we detected 280 (or some other small, deterministic number of) neuron connections in the human brain responsible for a drop-off in “high frustration” responses? Would you have a procedure to remove them performed on your unborn child if it was guaranteed to safely work, knowing they’d be a more mentally stable person? (Or, maybe more realistically, a drug that cured “mental illness” with zero side effects, same thought experiment. And please note I’m not saying that having or showing any emotion is bad, it’s what makes us human.) I mean, it’s possible in the machines that sound like us, apparently!
-
Comment on US regulator bans imports of new foreign-made routers, citing security concerns in ~tech
tauon Link ParentNo, unfortunately not, but I’m sure by now it’ll have been covered elsewhere as well. Tangent: I’m fairly certain this behavior of forced pay-or-allow-tracking will not be around for that much...No, unfortunately not, but I’m sure by now it’ll have been covered elsewhere as well.
Tangent: I’m fairly certain this behavior of forced pay-or-allow-tracking will not be around for that much longer with EU publishers, if it’s any consolation. Here’s an auto-translated article on the topic if anyone reading is curious (not tracking-walled :-)).
-
Comment on The Treasury just declared the US insolvent. The media missed it. in ~society
tauon (edited )LinkBrief, but stunning commentary piece in the Fortune. They make a great analogy that’s easy to remember: I was also really surprised by this: They can’t even admit the US government is running on...Brief, but stunning commentary piece in the Fortune.
They make a great analogy that’s easy to remember:
Most people cannot relate to trillion-dollar figures on a government ledger. So consider this: divide every number by 100 million — drop eight zeros — and federal finances look like a household budget in freefall.
That household earns $52,446 and spends $73,378 — running a $20,932 annual deficit. Its total liabilities and unfunded promises amount to $1,361,788 against just $60,554 in assets, leaving it $1.3 million in the hole. Uncle Sam, by any accounting standard, is insolvent.
I was also really surprised by this:
The Government Accountability Office (GAO) issued a disclaimer of opinion on the U.S. government’s FY 2025 financial statements — the 29th consecutive year it has been unable to determine whether the statements are fairly presented.
They can’t even admit the US government is running on borrowed time, money, and goodwill – and it’s been like that since the turn of the century!
One really does wonder if and when it all comes crashing down, US dollar and rest of the world included. -
The Treasury just declared the US insolvent. The media missed it.
38 votes -
Comment on US regulator bans imports of new foreign-made routers, citing security concerns in ~tech
tauon Link ParentGerman tech news outlet Heise has the answer: USA bans all new routers for consumersI'd be shocked if any major networking vendors actually manufacture routers and other devices in the US.
German tech news outlet Heise has the answer: USA bans all new routers for consumers
-
Comment on The first multi-behavior brain upload in ~science
tauon Link Parent100% we are barreling towards something difficult. “Ethically complicated” is the undersell of the decade. For some reasons as to why, I implore everyone curious enough to check out this thread in...100% we are barreling towards something difficult. “Ethically complicated” is the undersell of the decade.
For some reasons as to why, I implore everyone curious enough to check out this thread in the first place to read https://qntm.org/mmacevedo.
It’s essentially sci-fi horror written as a future Wikipedia-style article. -
Comment on Is it worthwhile to run local LLMs for coding today? in ~comp
tauon Link ParentInteresting! I’ll have to do some more testing, then. :-) I haven’t really gotten into the details with that yet so far as my hardware isn’t that great, so I’ll generally only play around with...Interesting! I’ll have to do some more testing, then. :-)
I haven’t really gotten into the details with that yet so far as my hardware isn’t that great, so I’ll generally only play around with local models, but not actually try to be productive with them, but I’ll keep this in mind.
I was especially surprised with this recent Qwen 3.5 release as they published both 27B and 35B models, two distillations so “close” together in a range above 10B parameters hasn’t really been usual as far as I can tell? And I might even be able to run them as 4-bit! -
Comment on Is it worthwhile to run local LLMs for coding today? in ~comp
tauon Link ParentI’d like to point you to some of the lengthy, very in-depth industry analyses of Ed Zitron’s blog to re-change your mind again :-) Most recently I started reading his recent NVIDIA analysis....wondering if the current AI boom really is a bubble
I’d like to point you to some of the lengthy, very in-depth industry analyses of Ed Zitron’s blog to re-change your mind again :-)
Most recently I started reading his recent NVIDIA analysis.
Genuine thoughts at the moment: The tech itself in the AI and LLM space can be truly incredible, and these companies’ valuations can be massively overblown currently. Both could easily hold true at the same time.
-
Comment on Is it worthwhile to run local LLMs for coding today? in ~comp
tauon (edited )Link Parent70B parameters will probably be a rather tight fit for 48 GB of VRAM at say an 8-bit quantization, based on the model size numbers for the newly released Qwen 3.5 family here. Edit: But that...70B parameters will probably be a rather tight fit for 48 GB of VRAM at say an 8-bit quantization, based on the model size numbers for the newly released Qwen 3.5 family here.
Edit: But that doesn’t mean the smaller models aren’t capable! I mean, I’ve tested both
qwen3.5:2bandqwen3.5:9bon nearly four year old fanless laptop hardware and was surprised at the quality of outputs.
Innovation in the space is going pretty crazy currently, IMO. Even if most any locally-ran models won’t compare to the big guns you can get from cloud hardware… especially once you start to factor in the price of things. Subsidized models like GPT 5.3 (Codex) or Claude as Opus 4.6 are hard to beat in that regard, at least from what I can tell so far. -
Comment on Google quantum-proofs HTTPS by squeezing 15kB of data into 700-byte space in ~comp
tauon Link ParentMan, just imagine the damage that’d be possible if some foreign state actor ever manages to smuggle an asset high up into the Cloudflare chain. They have consolidated so much power over ensuring...Man, just imagine the damage that’d be possible if some foreign state actor ever manages to smuggle an asset high up into the Cloudflare chain. They have consolidated so much power over ensuring the modern internet is, well, in a functional state, within just one company.
Noise, but I just have to show schadenfreude
Huh, I wonder who’s the washed-up lawyer now?