Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
gsuuon
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
20 ms
·
241.
▲
Ask HN: What are your favorite programming blogs?
38 points
by
gsuuon
3y ago
|
23 comments
242.
▲
by
gsuuon
3y ago
I think this is an important perspective - in the end you need to do what the company asks of you and not what you think is right for the company if the two are in conflict.
243.
▲
by
gsuuon
3y ago
The other problem is that it's not possible to request (as an extension author) or give (as a user) fine-grained permissions to extensions. You're only able to request very coarse permissions, often it's either you're gi
244.
▲
by
gsuuon
3y ago
I daily drive nushell - combined with wezterm it's great for a completely cross-platform setup. There are definitely usability and stability issues though, to the point where I wouldn't yet recommend it to anyone who isn't re
245.
▲
by
gsuuon
3y ago
There is a 'Coming to Nu' section of the nushell book which covers 'how do I do x' coming from other shells: https://www.nushell.sh/book/coming_from_bash.html
246.
▲
by
gsuuon
3y ago
Yup, I'm building a library that runs LLM's in browser with tagged template literals: https://github.com/gsuuon/ad-llama I think it has fundamental DX benefits over python for complex prompt chaining (or I wo
247.
▲
by
gsuuon
3y ago
As someone writing an AI library in Typescript - if AI is your main focus, learn Python. You'll have access to the largest AI libraries and frameworks with the most support materials and userbase.
248.
▲
by
gsuuon
3y ago
Awesome, thanks! I've been wanting exactly this service.
249.
▲
by
gsuuon
3y ago
This looks like it might be neat but it has a pretty sparse intro page and then the email signup goes straight into stripe checkout - is there any more info about the service anywhere? Like which models are available, or more pricing info?
250.
▲
by
gsuuon
3y ago
I think if you've spent more than 2 full weeks trying to figure out an effective note-taking system, it's time to just design and build out your own. Nothing will ever be more perfect for you than the one you build for yourself. M
251.
▲
by
gsuuon
3y ago
I always have a dumping ground of uncategorized piecemeal notes in whatever note-taker I'm using at the time which I usually call the 'heap' (like a heap of dirty laundry)
252.
▲
by
gsuuon
3y ago
My plugin is here: https://github.com/gsuuon/llm.nvim -- one of the "starter prompts" is commit message, so with vim-fugitive I open up the git status window, stage my changes, press 'cc', then 
253.
▲
by
gsuuon
3y ago
The F# for Fun and Profit[1] site by Scott Wlaschin is phenomenal as a general programming education site, not just for F# specific things (though there's plenty of that as well). The "Designing with types" series[2] is a goo
254.
▲
by
gsuuon
3y ago
Looks neat! A self-hosted and open source budget tracker is a great idea. The video skips the data import part, how does that work? Would it be possible to hook up to plaid?
255.
▲
by
gsuuon
3y ago
I think AI code assistants in-editor would be a really bad experience for beginners since discovering and dealing with hallucinations is going to be so much more painful. If someone is starting out, the best way to use AI to learn is probab
256.
▲
by
gsuuon
3y ago
It's tough but feels worth it - I never feel like I've missed out on anything with my daughter since I'm always around. I think if I can just structure my day better it'll all work out a lot better, just haven't fig
257.
▲
Ask HN: How are you using LLM's for code?
4 points
by
gsuuon
3y ago
|
3 comments
258.
▲
by
gsuuon
3y ago
How did you manage to keep a functioning brain with a kid and 4 hours of sleep a night? Were you able to sustain this long term or was this more of how the worst of it was? Were those 6 hours completely uninterrupted (and eating at desk)? I
259.
▲
by
gsuuon
3y ago
A lot of open source projects have Discord's - just pick a tech that you use and see if they link to one on their site or readme
260.
▲
by
gsuuon
3y ago
I think the value add of the speculative markets is (in theory) that people make predictions of what will be valuable -- if they're right they earn and if not they lose out. In effect, it's market research expressed as bets. That
261.
▲
by
gsuuon
3y ago
I've made something similar too (but as part of a neovim plugin) - it'd be helpful to be able to specify more information about the commit message, e.g. that it should conform to conventional commits.
262.
▲
by
gsuuon
3y ago
I'd never seen the comparison of BEM to OOP and utility-classes to functional programming - I can see how one is more focused on composability and the other extension. I'm not seeing how the approach given here compares with tailw
263.
▲
by
gsuuon
3y ago
I think the main concept introduced here is to hand-write what would be the match expression for each union. I've done similar stuff in languages with no ADT's, but without exhaustiveness checking you're left with co-locating
264.
▲
by
gsuuon
3y ago
> Coeffects are the dual of effects. Effects represent what your program does to the world while coeffects track what your program requires from the world. I think this is a neat follow up to both the how to read types[1] and the railway
265.
▲
Coeffects: Context-Aware Programming Languages
(tomasp.net)
3 points
by
gsuuon
3y ago
|
1 comments
266.
▲
by
gsuuon
3y ago
That's really interesting! I've wanted to do something similar but assumed that the token count required to ingrain a meaningful amount of information would make it very difficult. How many tokens do you typically need to feed int
267.
▲
by
gsuuon
3y ago
I'll use a debugger sometimes when the error is getting thrown in code I don't control but can view the source of - it can help figure out context and what the external code is doing. Otherwise I normally just put print statements
268.
▲
by
gsuuon
3y ago
Besides the three mentioned in the article (stringent accuracy requirements, fast edge inference, involved style transfer task) are there other good reasons to fine-tune?
269.
▲
by
gsuuon
3y ago
I think this is something that sample biasing would work better for, which you could do with local LLM's. For example with ad-llama[1] you would just have a sampler bias like so: const knownColumns = ['name', 'email&
270.
▲
by
gsuuon
3y ago
Really interesting to see the other side - I wonder if sometimes the perception of 'high context' vs 'low context' is really just 'not my familiar context' vs 'my familiar context'. > Excuse me, ma
More ›