selcuka

27 ms·

481.
by selcuka 1y ago
Dupe: https://news.ycombinator.com/item?id=45039661
482.
by selcuka 1y ago
It's been almost 30 years, so my memory is hazy, but that really looks like it. It still exists but last updated in 2000. It might be fun to make a Docker container with the right dependencies (if that's still possible) to run i
483.
by selcuka 1y ago
I also remember that the first web based forums used a Perl script that actually created new HTML files on the web server when someone posted a new topic, and modified the existing one when someone posted a reply. A bit similar to static we
484.
by selcuka 1y ago
Mozilla-based browsers used to (before Firefox 3, I believe) store bookmarks as HTML (bookmarks.html) instead of using an SQLite database. It still uses a single HTML file when exporting or importing bookmarks.
485.
by selcuka 1y ago
At least I have some numbers.
486.
by selcuka 1y ago
> they aren't being used by 80% of the world's population. I said "iPhone is more appealing to 80%+ of world's population". I didn't say everyone who wants it can afford it. It's pure speculation, of co
487.
by selcuka 1y ago
> Not surprising, this site is made for the Woz's of the world (and that's fine!). I'm pretty sure Woz perfectly understands why iPhone has a larger market share. I use a Pixel too, but I can see that an iPhone is more app
488.
by selcuka 1y ago
RFC 5785 (/.well-known/) breaks this convention, but sure, you are right in general.
489.
by selcuka 1y ago
Good point. I was thinking of an HTTP proxy, but surely a TCP proxy would work.
490.
by selcuka 1y ago
A DNS provider can not route your traffic through TCP proxies, so it must be the former.
491.
by selcuka 1y ago
> Unfortunately, guaranteeing anonymous internet porno is a terrible political beachhead to motivate "voters" to do anything. Reworded press release: "We protect children from being forced to upload their photos (on their
492.
by selcuka 1y ago
Yeah, "interfering with" is open to interpretation. One might argue that you don't have to say "hi" to yourself to interfere with, and it is sufficient to be within the same light cone.
493.
by selcuka 1y ago
Sorry, I meant that "implementing them wouldn't affect their bottom line in a negative way". I agree with you that they are useful features. That's what separates these bullet points from the others: Others will never be
494.
by selcuka 1y ago
Offline watching is a thing with streaming services, too. Also, there is no reason for a paid streaming service not to implement 1 (but not 1.2), 2, 3 and 4. It's not like these features will affect their bottom line. They just don&#x2
495.
by selcuka 1y ago
It could also solve the "where will the extra atoms to instantiate a second copy of yourself come from?" problem if the machine uses some kind of filament to 3d-print the traveller.
496.
by selcuka 1y ago
Doesn't any scenario of time travel imply an unlimited amount of entities? If you go back in time to observe (but not interfere with) your younger self, your younger self will get old and go back to that exact time too. So there will b
497.
by selcuka 1y ago
There is a 10-byte assembler (Intel) port of the same program: https://trixter.oldskool.org/2012/12/17/maze-generation-in-t...
498.
by selcuka 1y ago
Also their HID hardware was usually excellent. It's a shame they closed that division.
499.
by selcuka 1y ago
> A lot of (mostly non-US) orgs used locked-down managed IT and VMs where IE was still the only allowed browser, until the IE 11 shutdown in 2022, which is recent. That's hardly Microsoft's fault, isn't it?
500.
by selcuka 1y ago
> In the case of gpt-oss 120B that would means sqrt(5*120)=24B. That's actually in line with what I had (unscientifically) expected. Claude Sonnet 4 seems to agree: > The most accurate approach for your specific 120B MoE (5.1B ac
501.
by selcuka 1y ago
We are talking about accuracy, though. I don't see the point of MoE if a 120B MoE model is not as accurate as even a 32B model.
502.
by selcuka 1y ago
> GPT OSS 20B is a sparse MoE model. This means it only uses a fraction (3.6B) at a time. They compared it to GPT OSS 120B, which activates 5.1B parameters per token. Given the size of the model it's more than fair to compare it to
503.
by selcuka 1y ago
Note that these are only used for chat. As far as I understand there are no built-in system prompts when you use their APIs (or maybe they have different, smaller system prompts). I guess the rationale is that the end users of chat are not
504.
by selcuka 1y ago
> The most obvious way to adjust the behavior of a LLM is fine-tuning. Yes, but fine-tuning is expensive. It's also permanent. System prompts can be changed on a whim. How would you change "today's date" by fine-tunin
505.
by selcuka 1y ago
They are trying to create a useful tool, but they are also trying to beat the benchmarks. I'm sure they fine tune the system prompt to score higher at the most well known ones.
506.
by selcuka 1y ago
> Would a system with "no" prompt just spit out some random prompt? They claim that GPT 5 doesn't hallucinate, so there's that.
507.
by selcuka 1y ago
You are missing the point of my reply. Yes, your LLM can craft the syntax of the curl query required to make the API call, but how is it going to actually execute the curl binary? MCP is a standard method of doing something other than print
508.
by selcuka 1y ago
> if they could recreate the entire thing with the same tool "Within one year" is the key part. The product is only part of the equation. If a startup was launched one year ago and is worth $1B today, there is no way you can la
509.
by selcuka 1y ago
It shouldn't care about the format, true. But the LLM needs a mechanism to be able to connect to that tool from a sandboxed environment. MCP is the glue between the LLM and the actual tool. Technically you can expose a full HTTP proxy
510.
by selcuka 1y ago
I think it was a simplified example. The exact text is: > What emerged wasn't just a unique identifier. It was a compressed database record

More ›