gary's recent activity
-
Comment on What can be done about Linux being slopified? in ~tech
-
Comment on Anthropic says revenue jumped 45% to $16.8 billion in third quarter in ~tech
gary Link ParentMeh, every large tech company I've ever looked makes adjustments in their reports. GAAP doesn't fit very well with what these companies do. Investors are aware that adjustments are made; they are...Meh, every large tech company I've ever looked makes adjustments in their reports. GAAP doesn't fit very well with what these companies do. Investors are aware that adjustments are made; they are interested in different things and these presentations cater to the metrics they actually care about. I don't think it's likely that Anthropic would lie given that if they are angling for that IPO, all their numbers will come out shortly anyway.
-
Comment on AI models may be capable of self-replicating without access to a datacenter in ~tech
gary Link ParentI agree that revenue may not mean enough, but what it will tell us is what companies are willing to pay for this capability. If they have ~$100 billion in revenue one day, it tells us that there...I agree that revenue may not mean enough, but what it will tell us is what companies are willing to pay for this capability. If they have ~$100 billion in revenue one day, it tells us that there is a validated sizable market just waiting to be served at the right price. That's why I think the focus on ARR is misguided. On another note, revenue itself is hard to maintain. I mentioned on here a few months ago that my spending rate on AI was $800/month at work. It's now $350/month despite me using it more, because Opus 5.5 is just that much better that tasks get done more efficiently (or I got better at using AI).
-
Comment on AI models may be capable of self-replicating without access to a datacenter in ~tech
gary Link ParentLet's not shift the goalposts to profitability. The claims around here, and from you at least twice, are that the ARR numbers are smoke and mirrors not grounded in reality. I think if we get to a...I'll happily say "guess I was wrong" if the SEC filings show profitability remotely within the realms of their magic charts.
Let's not shift the goalposts to profitability. The claims around here, and from you at least twice, are that the ARR numbers are smoke and mirrors not grounded in reality. I think if we get to a financial disclosure that shows they had real revenue to back up their ARR claims (i.e. ballpark of ARR / 12 or 13 is accurate to what a real disclosure says), that's enough for multiple people to come out here and acknowledge they were wrong. And I will admit I was wrong if the opposite happens.
-
Comment on Ukraine struck Yandex data centers, Russian bots went silent—now questions are mounting in ~society
gary Link ParentFine, then what you're saying is the article is a giant nothing burger (computers that are down stop having activity) and wasn't worth posting.Fine, then what you're saying is the article is a giant nothing burger (computers that are down stop having activity) and wasn't worth posting.
-
Comment on Solo developer rebuilds Adobe Creative Suite in Rust using Claude, releases it free to all in ~tech
gary Link ParentThis? If you think that by not having a new model trained on the new languages/frameworks the existing ones can't write in the new languages/frameworks, you're mistaken. They can. And with tool...Because software development will continue to evolve. New frameworks made, new languages created, languages will receive updates, etc. Eventually, fairly quickly even models will end up being outdated with nobody updating them. That's why.
This? If you think that by not having a new model trained on the new languages/frameworks the existing ones can't write in the new languages/frameworks, you're mistaken. They can. And with tool calls, they can do a lot in general (humans can still write the tools easily, like LSPs are human-written after all). There's also fine-tuning that can be done by non-frontier labs. There are people who do it at home as a hobby.
-
Comment on Solo developer rebuilds Adobe Creative Suite in Rust using Claude, releases it free to all in ~tech
gary Link ParentWhy would those models no longer be usable? There is no reason to believe that these models will suddenly stop being runnable on hardware. No, because the models are open weight. You can read the...If they are no longer provided to the world for free you either end up with models that eventually will no longer be usable
Why would those models no longer be usable? There is no reason to believe that these models will suddenly stop being runnable on hardware.
because interference providers are now required to start paying for these models.
No, because the models are open weight. You can read the license if you want; the models are runnable forever for these inference providers. If future models have a license change, that does not retroactively apply to what was already released.
-
Comment on AI models may be capable of self-replicating without access to a datacenter in ~tech
gary Link ParentITAR can't rein in the Chinese, who are a close second in training. It can slow them down by removing Nvidia exports, but as we're seeing with the Huawei Ascend line, it's only a matter of time...ITAR can't rein in the Chinese, who are a close second in training. It can slow them down by removing Nvidia exports, but as we're seeing with the Huawei Ascend line, it's only a matter of time before they get good enough with hardware for that to no longer be a barrier.
-
Comment on Solo developer rebuilds Adobe Creative Suite in Rust using Claude, releases it free to all in ~tech
gary (edited )Link ParentThat literally doesn't match the point we were discussing at all, which is the cost of using models. The discussion was about if the current pricing can be rug pulled. EDIT: Qwen 3.8 is an open...That literally doesn't match the point we were discussing at all, which is the cost of using models. The discussion was about if the current pricing can be rug pulled.
EDIT: Qwen 3.8 is an open weight model. We know the cost to serve it is X/million input tokens and Y/million output tokens. And a cached pricing per token. It does not matter if it cost a trillion dollars to come up with the model, if the model weights are now open and we know that it's a static X/million input and Y/million output. That X and Y do not change except if the cost of hardware/electricity changes with inflation. Each frontier open weight release effectively becomes a floor on the cost-to-quality code output we can have. It will never get worse than what it is today, accounting for inflation.
-
Comment on Ukraine struck Yandex data centers, Russian bots went silent—now questions are mounting in ~society
gary Link ParentEverything that is "misleading" involves assumptions, yeah. That's a tautology. And every facet of communication involves assumptions. We can't hide behind weasel words and technicalities when we...Everything that is "misleading" involves assumptions, yeah. That's a tautology. And every facet of communication involves assumptions. We can't hide behind weasel words and technicalities when we communicate lest every single statement needs to be done with every possible caveat. Given the discourse surrounding this topic from the rest of the internet, it's very much clear this title was clickbait to imply it was Russian social media bot traffic that dropped.
-
Comment on Ukraine struck Yandex data centers, Russian bots went silent—now questions are mounting in ~society
gary Link Parent"Went silent" and "now questions are mounting" imply this was a drop in nefarious activity. Nefarious activity would be Russian bots on social media, given the context of the last decade. The..."Went silent" and "now questions are mounting" imply this was a drop in nefarious activity. Nefarious activity would be Russian bots on social media, given the context of the last decade. The "went silent" part also implies a communication angle. Then you open the article and it's all about search engine crawlers which are not nefarious.
-
Comment on Solo developer rebuilds Adobe Creative Suite in Rust using Claude, releases it free to all in ~tech
gary Link ParentWhy not? Let's pretend tomorrow Anthropic, OpenAI, and all other labs implode financially and governments ban all new training. The model weights still exist. Data centers could run the existing...Why not? Let's pretend tomorrow Anthropic, OpenAI, and all other labs implode financially and governments ban all new training. The model weights still exist. Data centers could run the existing models at a low cost. I could still churn out hundreds of thousands of lines of code in a day at a relatively low cost.
-
Comment on Solo developer rebuilds Adobe Creative Suite in Rust using Claude, releases it free to all in ~tech
gary Link ParentDoesn't matter if the cost of training new models is high and that is subsidized. If humanity never trains a new model ever again, we will still have Astra/Opus/Kimi/GLM level models to use at low...Doesn't matter if the cost of training new models is high and that is subsidized. If humanity never trains a new model ever again, we will still have Astra/Opus/Kimi/GLM level models to use at low cost for the rest of time. Thus, the cost of code is low even if the cost to train models is high. The latter is irrelevant to the former.
-
Comment on Unofficial Linux app for music streaming service Qobuz QBZ shut down after receiving cease and desist letter in ~music
gary Link ParentQobuz doesn't give a larger share; they operate on pro rata so unless they negotiated some weird terms, they're doing the industry standard of ~70% revenue. You can give your $10 to most music...Qobuz doesn't give a larger share; they operate on pro rata so unless they negotiated some weird terms, they're doing the industry standard of ~70% revenue. You can give your $10 to most music services and $7 goes to the label. Roughly. Spotify's trying to do some shenanigans with audiobooks, but it's still playing out in court. Qobuz has some slick marketing to make it sound like they're giving a higher share, but they're not.
A service can look like it pays out a higher amount per stream if they don't have free trials, if they don't have ad supported plans, or if their user base just streams less for some reason. All ways you can change the dollar/stream number. But if the worry is how much of your dollar goes to the label versus the platform, it's 70% just about everywhere.
-
Comment on Advice on TV and media storage in ~tech
gary Link ParentRoku has the same Automatic Content Recognition that the privacy-sucking TV smart OSes do. Do not get a Roku if that's a concern. The only good choices are: Linux box you build, a trusted Android...Roku has the same Automatic Content Recognition that the privacy-sucking TV smart OSes do. Do not get a Roku if that's a concern. The only good choices are: Linux box you build, a trusted Android device (idk what are), or an Apple TV.
-
Comment on Given everything we know about AI so far, what do you think it can or cannot be good for? in ~tech
gary Link ParentAs a semi-watch guy, the example might end up proving the opposite. There's some hardcore fans of G-Shocks, Timex, Swatch, etc. I fell in love with G-Shock last year coming from mechanicals and...As a semi-watch guy, the example might end up proving the opposite. There's some hardcore fans of G-Shocks, Timex, Swatch, etc. I fell in love with G-Shock last year coming from mechanicals and it's all I wear 99% of the time now. Within the mechanical watch world, the majority are cranked out by the millions in a factory at the low end (Orient), medium end, and the relatively high end (see Rolex). But they're all still loved and can sustain an artificial scarcity model at the high end. Sure, at the highest end, maybe they're handmade, but the supermajority of the market of mechanical watches are not handmade.
-
Comment on The return of phones with physical keyboards? in ~tech
gary Link Parent73 WPM on 98% accuracy (monkeytype.com) using iOS touchscreen without the word autocorrect kicking in, just the guessing-on-letter logic. I think my thumbs would hurt if I tried that on a hardware...73 WPM on 98% accuracy (monkeytype.com) using iOS touchscreen without the word autocorrect kicking in, just the guessing-on-letter logic. I think my thumbs would hurt if I tried that on a hardware phone keyboard. I used to have a Nokia with one and strained around the 40 WPM. But the tactility of a physical keyboard is nice...
-
Comment on Saudi Arabia is ready to take back Yemen. Can it be done? in ~society
gary Link ParentIt's not Saudi forces and even less so Pakistani ones. It's the Yemen government forces with Saudi air support and probably Pakistani intelligence support. In terms of troops and readiness,...It's not Saudi forces and even less so Pakistani ones. It's the Yemen government forces with Saudi air support and probably Pakistani intelligence support. In terms of troops and readiness, Pakistan hasn't been impacted much here if at all. They might join later, but it's premature to characterize it as "Saudi/Pakistani forces" at the moment.
By the way, this Simon Whistler guy is a front for a bunch of slop channels. He's just reading what ghost writers/content mills/potentially AI is churning out; there's no way he's fact checking it all.
-
Comment on Alright, ladies and gentlemen. Thinking caps needed. #Apple #movies #tv in ~tech
gary Link ParentThe only existing solution I'm aware of is Apple TV, yes, but there's potentially a product worth building here that's an alternative frontend for browsing an Apple TV library. @moocow1452 also...The only existing solution I'm aware of is Apple TV, yes, but there's potentially a product worth building here that's an alternative frontend for browsing an Apple TV library. @moocow1452 also had a good suggestion with Movies Anywhere since it allows building custom playlists, but that only covers ~80% of a user's library from my experience and 0% of tv shows since MA is a movie service only.
-
Comment on Alright, ladies and gentlemen. Thinking caps needed. #Apple #movies #tv in ~tech
gary LinkGuys, Whitewatermoose's purchases are through Apple TV/iTunes. Think "digital rental on Apple's streaming platform with DRM and no known expiration time" rather than "digital file that can be...Guys, Whitewatermoose's purchases are through Apple TV/iTunes. Think "digital rental on Apple's streaming platform with DRM and no known expiration time" rather than "digital file that can be copied and pasted". Hence, Plex, Jellyfin, and ersatztv are cool software but don't apply here.
Yes, from that example, your coworker is clearly bad at prompting if they're rewriting SSH and asking you to review 11k lines. The tool might have output something of poor quality, but if your coworker couldn't detect that before asking for a code review, then what faith should one have in their ability to use the tool? For your own experience, if you find that Codex is doing more than you wanted it to do, then describe what you think it should have done instead and see if the next time, the prompt being more descriptive leads to the outcome you want. Or better yet, describe what the code change should have been, ask Codex to diff it, then ask Codex how the prompt could have been worded and store that difference in your expectations vs Codex's expectation in AGENTS.md.