padolsey

11 ms·

151.
by padolsey 2y ago
I now think of single-forward-pass single-model alignment as a kind of false narrative of progress. The supposed implications of 'bad' completions is that the model will do 'bad things' in the real material world, but if
152.
by padolsey 2y ago
This study just seems a forced ranking with arbitrary params? Like, I could assemble different rules/multipliers and note some other cooperation variance amongst n models. The behaviours observed might just be artefacts of their specif
153.
Show HN: xmllm – Structured LLM streaming output using lenient XML parsing(github.com)
5 points by padolsey 2y ago|0 comments
154.
by padolsey 2y ago
> communicating complex ideas in conversation is hard and lossy True but.. > instead of building the website, the model would generate an interface for you to build it, where every user input to that interface queries the large model
155.
by padolsey 2y ago
>exceptionally smart at making predictions Is anyone though? Genuine question. I don't have much faith in predictions anymore.
156.
by padolsey 2y ago
Was it just me who thought that this was _already_ how LLMs worked? I'd always assumed they were -- so to speak -- swimming in their own embeddings space before coming out on the other side with language. But it turns out they're
157.
by padolsey 2y ago
I imagine this is a much harder problem than meets the eye. Even for my own travel, there are like a dozen things I'm trying to optimize for. Time, money, seat, travel to/from airport (traffic, train cost, taxi pickup/dropoff
158.
by padolsey 2y ago
So basically this just adds random characters to input prompts to break jailbreaking attempts? IMHO If you can't make a single-inference solution, you may as well just run a couple of output filters, no? That appeared to have reasonabl
159.
by padolsey 2y ago
I was thinking that. And really chuffed to see things like $(function(){ $(".rollovers a").hover(function(){ $(this).children("span").fadeOut(); }, function(){ $(this).children("span&
160.
LLM Security: Keep Untrusted Content in the User Role–Always(blog.j11y.io)
1 points by padolsey 2y ago|0 comments
161.
by padolsey 2y ago
> Programmatically assemble prompts for LLMs using JavaScript. > $`Analyze ${env.files} and report errors. Use gitmojis.` This is kinda misleading and confusing as a lead. I could literally say: > Programmatically ass
162.
by padolsey 2y ago
The fundamental flaw people make is assuming that LLMs (i.e. a single inference) are a lone solution when in-fact they're just part of a larger solution. If you pool together agents in a way where deterministic code meets and and verif
163.
Saizeriya's Secret to Survival(hiddenjapan.substack.com)
1 points by padolsey 2y ago|0 comments
164.
by padolsey 2y ago
Umm it does have support for open/free/self-hosted models through either openrouter or by changing the oai endpoint (as long as you've got an openai compatible endpoint which is usually the case).
165.
by padolsey 2y ago
I was in in the `concat->copy->[Claude/OAI]->paste` workflow until a few weeks ago. Cursor is a revelation. Nowadays I only very rarely pump things wholesale into Claude[dot]ai when I'm dealing with a massive cross-codeba
166.
by padolsey 2y ago
Cursor is a fully fledged fork of VSCode so IMHO it's in a different ballpark to Aider/Claude-Engineer/etc. Its LLM assistance is baked in to the IDE. Feels really comfortable. I switched over from SublimeText after a decade.
167.
by padolsey 2y ago
Cursor is incredible. You can use whatever model you like. Give it your openAI key if you want o1-preview. Using it with Claude Sonnet is usually enough though. I use it everyday and it makes it possible to work between dependencies and it
168.
by padolsey 2y ago
I haven’t tried, and tbh I’d be very scared of messing with the complete cocktail of meds (~7) I’m on as it is. Risky! I’d rather maintain a knowable stability even if it’s non optimal.
169.
by padolsey 2y ago
This is such an interesting thing to read about!! Thank you for posting it! I have epilepsy following hemmorhaging to the parietal area affecting my motor movement, planning, and sensation. Much like the patient described, specific things i
170.
by padolsey 2y ago
It's BrailleInstitute, an org catering to those with visual impairments. Understandably, their site is accommodating to their core audience.
171.
by padolsey 2y ago
Niiice! I really like it. The spatial approach is cool, though labelling/annotations/axes would help. I share the frustraion with getting book covers for my project ablf.io. Amazon used to make this much easier, but they've l
172.
by padolsey 2y ago
> gives me the information I need. I mean, I'd like at least a brief blurb about their entire premise of safety. Maybe a definition or indication of a public consultation or... something.. otherwise the insinuation is that these thr
173.
by padolsey 2y ago
Technically it's acting on behalf of a proactive user in Chrome so IMHO is non-"robotic". But heh tbf this was also the excuse of Perplexity where they argued they are a legitimate non-robotic-user-agent (thus don't need
174.
by padolsey 2y ago
Yeh sure - what did you have in mind?
175.
by padolsey 2y ago
I really love this kind of thing. I made one called 'redoku' a long long time ago. It generates random challenges. Try it here: https://padolsey.github.io/redoku/
176.
by padolsey 2y ago
I've found Claude to be way too congratulatory and apologetic. I think they've observed this too and have tried to counter it by placing instructions like that in the system prompt. I think Anthropic are doing other experiments as
177.
by padolsey 2y ago
> If we're talking psychoanlytically of where wanting comes from, and call it a desire to fill a void of how incoherent you actually are, then an LLM doesn't go through that process. Ironically, many people complain LLMs are to
178.
by padolsey 2y ago
> No matter how fast Searle is, he won't be able to come up with a beautiful and original Chinese poem that has the creative spark special to humans Why not? > Of course, at some level of complexity, it will be stuck in a local m
179.
by padolsey 2y ago
Looks cool. To help the pitch, I think you should show the equiv price on anyscale/together for the 10B tokens. Also is there a reason you're not selling at more granular amounts? Why $400? Just speaking personally, I reckon I&#x2
180.
by padolsey 2y ago
I don't much mind the populurization of these terms or making people think in more novel ways. Even if it's seen as BS by those more informed, I'm glad such books are written and communicators like Taleb exist. Without him, I

More ›