I don’t get it. One day they tell us that AI is the most dangerous and the most advanced tech the humanity ever invented. Now we consider an environment with just a package proxy between it and the outer network a good enough effort to sandbox. I do see some contradictions. Considering they also effectively test next-gen models there, I think the only proper sandbox would be a physically separated network. You need packages, well, bring them with USB stick
I have no idea how we do it, but I think we should somehow ban the expansion of all the existing unions (EU, CIS, NATO, whatever else we got). The overall worldwide tension is pretty high, but now
- US goes after South America and Greenland
- EU goes after Canada and Georgia (country), which is just ridiculous in terms of geography
- Russia invades Ukraine and goes after Eastern Europe and Middle-East to some extent
- China does its expansion too
Even though some of these actions are just political nonsense, it still increases the overall level of escalation. This is not what the world needs right now
Trade agreements and unions — yes, please; yet another political or military union expansion — please, God, no. I understand that EU is a mix of both, but also do think that this is a purely political game against Trump
You dont really see a difference between US wanting to take over SA and Greenland militarily and Canada and Georgia wanting closer ties to the EU voluntarily?
You sound like Puting fan boys arguing about NATO expansion, NATO does not expand. countries voluntarily ask to join them, because the alternative is being destroyed
I think you miss the symmetry part. You like NATO, OK, that’s good. But will you be equally happy if China will fund PTO (Pacific Treaty Organization, no pun intended) and Mexico, Cuba, Algeria, Egypt, Brazil, etc., will absolutely voluntarily join it?
This can work if one country/alliance is doing it, but when everyone starts doing the same it becomes a zero sum game with the dead end. It’s pretty much the prisoner’s dilemma: “if we don’t expand, they will expand/invade” although it’s a common interest to not escalate
I’m an American, and I don’t see it as a purely political game against Trump—my country has dramatically shifted its position toward its allies from reliable partner for 80 years to aggressive antagonist both in trade and military. And it’s not just a rogue president—a full third of the US electorate supports these actions.
It makes sense to me that these countries would want to build new trade and security alliances with countries who have similar interests and who face similar threats.
The alternative to these defensive military pacts is that Russia or the US or whomever can literally invade and annex countries piecemeal to expand their power. If we can’t effectively ban that, then it doesn’t make sense to me to ban the only thing that deters it today.
I got X3 because it was on the ads all over my internet. My idea was that I can stick it to my phone when I go out or commute and don’t want to bring my standard Kobo reader, so I can read more instead of doomscrolling
So far, my impression:
- it’s a fun piece of tech, but it’s also a super simple one. Idk the price of a eink screen like that, but it’s basically a screen + esp32 + battery, so $70 might be a bit too much, and just includes all the ads budget, which is clearly huge
- CrossPoint firmware is neat and OSS, so you can for and DIY whatever you can imagine
- it’s still so much more convenient to read on a bit bigger reader. Even my Kobo Clara BW with Koreader, which is still quite portable, has way better UX. And it’s just about x2 more expensive
- magnet is not strong enough to sit firmly on my phone
- I got a epub rendering issue with one of my books
- I don’t read a lot on it and still prefer my Kobo, so it failed the main objective for me personally
- it’s still fun how much esp32 in such a tiny enclosure can do :)
I was looking at $10k Mac Studio with M5 Ultra and 256 GB for local experiments, but then struggled to find what really good modern model I can fit into it. Yes, it can run a good dense 27B at Q8 with plenty of context, but what beyond that? IIUC, some Deepseek flash variants at Q4 are also feasible, but I am not sure if the quality will be good. They also don’t run that fast, like about 30 t/s
So if I stay within 35B, especially MOE, my M5 Pro 64GB MBP can also run them well, and it can do plenty of other stuff too including gaming. While 256 GB with such RAM bandwidth and powerful GPU sounds like fun on paper, it doesn’t seem to be the next level compared to 64 GB
I feel like for localAI t/s is less of an issue. Just make a PRD and run a ralph loop. For big slogging projects like reverse engineering, or converting a codebase to a new language it actually doesn't matter if it takes a day or seven days.
Yeah this is my experience. My 24GB 3090 + 64GB RAM takes a couple hours to crank out some code with largest Gemma 4 and Qwen3.8 models it can run
But in the meantime I get dishes done, vacuum, flip laundry... etc etc
Frontier models also seem in such a rush to emit anything they produce a mess that needs steering all day anyway
While I have not tested it, it feels like my local setup going slower is better at producing code that works the first time as its not trying to look fast for marketing sake
Is Apple still doing the extended return window for items receiving around the holiday? I know sometimes they allow items received near the end of November to be returned in January, so in theory if you time your early order right you can use it, test it, and even have some time to play with it before you return it.
I haven’t had a chance to compare side by side, but people say that if you primarily read books/prose, the BW version is better because the screen is more crisp compared to the color version, which is still a bit blurry when you read books. That’s why I ordered BW. Did anyone compare these two side by side in real life for simple text?
I'm happy with my Kobo Clara Colour. As far as the specs go, both are 300 PPI when displaying black and white. I took a picture of my screen from as close as my phone would allow - the lens was basically touching the e-reader [1]. My eyes can't really focus at this distance anyway, so I can't complain. I don't use the colour screen much, but seeing the book covers in colour is fun for me.
It is maddeningly impossible to get an ereader with all three of these features: b&w screen for maximum crispness; physical buttons for reliable page turns; modern processor for speed.
It runs Android. It's a Chinese company and comes with a bunch of pre-installed apps, but I just blocked its abilities to phone home and use Koreader. It's been great for me
https://goodereader.com/blog/product/onyx-boox-go-7-bw
"The Onyx Boox Go 7 is a dedicated e-book reader with a black and white e-paper display. This device has the same specifications and software as the Go 7 Color Gen 2, except that this model doesn’t have a Kaleido 3; instead, it uses the latest generation Carta 1300 e-paper panel, which increases responsiveness. The key selling points of this model are the physical page-turn buttons and Google Android 13, as well as full access to the Play Store. For those of you who want an e-reader that doesn’t lock you into a specific ecosystem, this deserves a look."
I personally use a Moaan 7. Android and does everything i ask it (I just use FDroid apps though). But do bear in mind it doesnt seem to run apps in the background at all. Just whats onscreen. Battery life is extremely long (probably not unrelated)
Yeah thinking about an pocketbook era lite but apparently can feel a bit sluggish. Might have to unfortunately wait and hope for kobo to release something along the lines of a vision bw.
I rage quit the Boox because the page turn buttons were the same for both landscape and potrait, which meant that one direction was illogical. There were user patches but it just made me so annoyed that when a friend expressed interest in mine, I just gave it way. Maybe it's fixed now but so damn infuriating.
I use a bw Boox device, a colleague has the colour version of the same device. The colour version does not show text more blurry, rather, the contrast is significantly lower. I personally prefer the bw device.
> Did anyone compare these two side by side in real life for simple text?
Yes, I have - I have a Kobo Libra 2 (BW), Boox Go 7 II (Color - same screen as color Kobos) and Boox Note Air 3 (Color).
The issue isn't crispness really - when rendering B&W content the crispness is about the same. However, there's a massive issue with contrast - the Color screen background is very dark and they're basically unusable (for me) without driving them constantly at 80-100% backlight (even during the day). Even at 100% backlight, the page background is darker than the BW reader at like 30% backlight.
This makes the reader eat more battery while being less pleasant for text reading. Moreover, in dark mode (light text on black background) the color screens all show severe ghosting artifacts which don't appear as much on BW screens.
Having said that - reading comics and color books (e.g. programming books) looks great on the color screens. There's just something about that slightly less vibrant "papery" look that's very attractive.
I have a Libra 2 and a Libra Colour. I generally don't notice a difference in crispness, though the effective resolution of the colour's screen is lower. There is a huge difference in the contrast, and you will probably want to use the frontlight even in a reasonably bright environment.
Yes, I replaced a B&W Sage with the Libra Colour, and used them for a bit side-by-side. Contrast (not sharpness) of B&W text is slightly worse with the color one, but not by so much that you'd notice it if you weren't comparing them next to each other. The resolution of actual color is worse, but obviously you can't get color at all on the B&W one, so hard to argue that this is a downgrade.
Unless you're a real contrast freak, I think most people would be slightly happier with the color than the B&W; but it's definitely not worth paying extra for.
The C is sharper but the L looks more like a real book/paper. I believe as an artifact of the colour screen and it’s called “screenshot effect”. In the end, once you use either for 10 minutes it becomes “the new normal”.
I compared. On the color version, the text seems to have a slightly fine mesh or rippled texture. There's no huge difference, but I prefer the black-and-white version
I had a Kobo Libra 2. It broke and I replaced it with a Kobo Libra Colour. The biggest difference is that the Libra Colour is unusable without the backlight turned on, since the screen is so dark. The Libra 2 had a pretty white background without the backlight, whereas the Libra Colour has a grey background. I don't know if the Clara Colour also has a really dark screen or not.
It's the same, just a smaller 6" Kaleido 3 screen VS. 7". Personally I find the Libra Colour fine in a lot of conditions without the front light. Kaleido 3 is darker than BW screens but it's like bad newsprint vs. good newsprint.
> A prompt implying an image is present doesn't mean one is (the person may have forgotten to upload it), so Claude checks for itself.
Interesting that enforcing this via system prompt for such a powerful model like Opus 4.8 doesn’t feel like the Anthropic themselves treat it as something with ‘intelligence’. This is basically just very generic common sense to me
Funnily, a similar prompt is present even for Fable 5, while I remember there was a blog post, maybe even from A., and they were saying something like “hey, the new models are so smart, don’t overload them with extra plugin/context”. Well, they clearly aren’t. Don’t want to sound like an AI-skeptic, I use it daily, just stating the fact.
> Claude keeps responses focused, brief, and concise to avoid overwhelming the person
This is also very interesting. It pretty much ignores it by default. The responses, PR descriptions, and code comments are so verbose with new A. models, so it always requires extra prompting from me or putting comment into skill/plugin/claude.md to make them of a reasonable length
I’m also curious how it really ’weights’ all the instructions coming from main system prompt, my system prompt, skills/plugins, CLAUDE.ms, and nearby code/comments/readme. It clearly should follow some reasonable hierarchy, but because the model itself is so complex, I think (and it feels like) that there is such a mess in its context and reasoning. It deals with it surprisingly well, though, but wonder if it can be done in a more efficient way
For me, Claude usually says ``I don't know'' as first or second answer and stops with this ultra-concise word count of four or less.
(Answer number one before that is usually "I don't have internet access, from memory it is either A or B, but I cannot recall what you want to know." ChatGPT or Gemini can often do the search, while google.com AI assistant or perplexity just tell blatant lies. Copilot.com can do the search, but external links are invalid made-up stuff for harder questions, which seems to be the case 9 out of 10 times.)
Which is great, since it could answer with made-up BS, but does not.
AI, except for doing better web searches for a year now, hasn't really improved for my tasks in the last three years, except for coding. Then again, AGI benchmarks seem to go through the roof only above Sonnet 5 and self-hosting, so perhaps the questions I ask not too hard for long now.
And self-hosting, eve 1bit/1.5bit models are a pondering a little too long to comfortable run in summer, but cheap on the RAM and insanely good at coding since a month now all of a sudden.
I almost wonder if Claude reads that it “keeps responses focused, brief, and concise” and interprets that as built-in behavior and concludes that it doesn't need to spend additional effort enforcing it, just as it doesn't need to expend effort being “accessible via this web-based, mobile, or desktop chat interface”.
Claude doesn't "interpret" anything. That's not how LLMs work. They are still sentence completing machines. You give input in, the model network fires the chain that produces the most likely related output. That's it.
I think you might be right, though that this would have influence on other output tokens since it is phrased as a substantive property, rather than an instruction.
I suspect it's related to EU-required watermarking:
> Having said that, in areas where there is an arbitrary choice
between particular words or terms within the code, the
watermark can be used, such as comments within code. But by
definition, it will have a negligible effect on the actual code
produced. [1]
I've found that you can reduce the prolixity before (or afterwards) by asking it to use ASD-STE100 Simplified Technical English or ISO 24495-1 after deciding whether a particular comment is worthwhile. (I've since included a standard for "worthwhile" because the clanker doesn't have a good bead on _that_ matter.)
It's not that surprising if we remember that the model is trained to be a generically useful next-token predictor, not necessarily an agent or a chatbot. It needs to know about the environment it's embedded in and what assumptions it can make, and by design the only way to get that information in there is to put it in the system prompt. It's also possible that even if it could figure something out on its own, it's just more efficient to bake it in rather than having it dedicate attention and tokens to it on every prompt.
> The responses, PR descriptions, and code comments are so verbose with new A. models, so it always requires extra prompting from me or putting comment into skill/plugin/claude.md to make them of a reasonable length
I have mentioned this here before, but the majority of my organization has reacted viscerally to this verbosity that LLM-text has been forbidden: in comments, in PR/commit messages, in correspondence, in Jira tickets.
A couple non-coders who want to make PRs without writing the description are now rebelling and saying this can be fixed if we spend our time writing skills for Claude so it becomes readable again.
> This is basically just very generic common sense to me
It's important to remember that we are talking about a calculator that doesn't have an understanding of common sense. Unironically, this is common sense.
Yes, but I write this putting an ‘average AI company CEO’ hat on. We hear statements about outstanding intelligence (not just usefulness as a tool, which is no doubt already there), so it’s interesting to see that the authors themselves don’t treat it like that
I think the old Dijkstra quote apples now more than ever:
“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.”
Whatever these things are doing, it’s not the same as what a person does. Trying to decide if whatever they do fits into the box we label as “intelligence” is completely uninteresting, in my view. What’s interesting is figuring out just what they can do and how best to use them, which sounds like a related question but really isn’t.
I’m a huge fan of Mozilla and Firefox specifically, but I don’t think that the way classical adblock extensions are made is the right way. Instead, it should be done as Apple/Safari do it [1]: the browser provides an API to hook/set a block list of identifiers that should be blocked, it could be resource hostnames, html signatures, need to think how to improve the API, but this way it’s completely safe, extension has zero access to the actual page content. Apple does the same for caller id apps. Afaik, Android has this API too, but the last time I checked, all relevant extensions were just “give me your whole phone control or web page access”, so Google clearly doesn’t enforce it
Yes, it kinda gives more control to the platform, but so far, iOS extensions that use this API worked surprisingly well for me
Please, correct me if I am wrong and uBclock can already work in this restricted mode
And the last thing, if people really want to give someone a full page content access, they surely should be able to do that, so kudos to Mozilla
Absolutely not. This is exactly why people have a problem with Chromium, which now has some of the same restrictions on extensions as Safari (Manifest V3).
One of my biggest annoyances about my iPad is how much less effective the content blockers on Safari are than on real Firefox and ublock origin that I have on my Android phone and had on my previous windows tablet. Also iOS “Firefox” can’t run any adblockers, whether ublock origin or iOS content blockers so I have to use Safari and put up without consent-o-matic or tab sync
For example, ublock origin is able to patch out anti-adblock scripts that Safari content blockers cannot. Try tvtropes for an example. Works fine with firefox and ublock origin, displays a “allow ads of subscribe” instead of the content on safari.
Kagi's Orion browser is a far as I know the only one on iOS that supports proper Firefox extensions, including ublock and others. You may want to give that a try.
You can only get a very limited uBlock Origin facsimile, not the real thing. A lot of browsers have a uBlock Lite. It's missing a lot of useful features IMO.
So you want to make it super easy to bypass the block by randomising identifiers every so often, and then the blocker will have to go through the full update process?
Maybe this works for some extensions but an ad blocker has to be maximally dynamic to work.
With that approach how am I going to block ads on long-tail-site.example.com and lots-of-ads.new-domain.com? I'll let uBO read my traffic. Firefox is reading my traffic anyway.
