sh3rl0ck

I've been a strong proponent of reallocating all LinkedIn server capacity to GitHub.

Everything that went wrong with Claude so far.

There's no mention of SLMs or LLMs, though.

This work represents a compelling real-world demonstration of “tiny AI” — highly specialised, minimal-footprint neural networks

FPGAs for Neural Networks have been s thing since before the LLM era.

One of the very few good things from the AI race has been everyone finally publishing more data APIs out in the open, and making their tools usable via CLIs (or extensible APIs).

I assume it's an economies of scale thing now.

It's not like Apple is putting any thought into either the UX or the engineering side of utilising the compute properly (except calculating those glass effects extra inefficiently).

Minimise SKUs and get some use out of the binned chips who have a few failed cores.

I think there's a reward for finesse too.

As you mentioned, scope definition and constraints play a major role but ensuring that you don't just go for the first slop result but refine it pays off. It helps to have a very clear mental model of feature constraints that doesn't fall prey to scope creep.

In Gallifrey? In Gallifrey.

Never had Nvidia issues on Fedora and Ubuntu so far, 1P a multi computer research lab as well.

I find this a tad funny since ccc is my claude code alias, since cc is taken up by the actual, working, greatly optimised and really well made Clang C compiler.

It doesn't "require" a launcher at all; but the people have the freedom to change theirs.

Kinda like MacOS users only have Aqua, whereas Linux has a lot of DEs, the choice of which is entirely handed to the user.

The poor maids and servants, the poor chauffeur, the poor chef, etc.

'Before you joined' seems to that she doesn't anymore.

I find it a little amusing that AI companies provide free AI subscriptions to their employees and their families. Perhaps because I'd never thought of it that way.

That's kinda nice actually; I assume employees get early access to models to test (5.3 codex, for example). Do families get it too?

Should be, but sadly isn't; the incentive is to feed into the economically incentivised "bs" rather than be wary of it, which is just... disheartening. Schools often inculcate this enough on a subconscious level that I've seen my fellow college mates rather unable to shake off that residual behaviour.

My problem with AI support agents is that instead of being good classifiers for my problem, they're even more non deterministic now in terms of grievance or query redressal. If a listed option is not available, I'd rather talk to a human who'd know or have the ability to find out, instead of an LLM which would just make something up with no grounding.

Yeah, I feel Anthropic is just very deliberately theatrical about the way they present their technology and company and even how they price them. Dario's conviction seems too over dramatic to be real to me, but while there's a chance he's drinking his own kool-aid, they just know how to present it as a premium experience, and their developer adoption helps with that.

I'd wager that lower tok/s vs lower quality of output would be two very different knobs to turn.

On one hand, with the top comments of the rebrand post showing how many insecure deployments there are, something like this alongside cloudflare zero trust is probably a much more secure solution.

On the other hand, I just wanna point out

Firstly, Cloudflare Workers has never been so compatible with Node.js. Where in the past we had to mock APIs to get some packages running, now those APIs are supported natively by the Workers Runtime.

Deployed a project a couple of days ago, and compared to past attempts where I had to wrangle (pun intended) with certain configs for deployment styles for node based applications, the normal build tooling just worked out of the box. Planning to move a couple of my free-from-me high DAU user projects that are on the vercel premium tier over to CF workers.

What would you categorise as a "cool looking job"?

I'm 20 and I read Winamp, ha.

I reckon that the majority of the HN audience is into older (perhaps more graceful) working technology despite all the fanfare that newer, shinier stuff garners.

Beyond Linting and Shell Exec (gh, Playwright etc), what other additional tools did you find useful for your tasks, HN?!

Most of my feedback that can be automated is done either by this or by fuzzing. Would love to hear about other optimisations y'all have found.

Is it weird that I now know exactly which xkcd it will be just with conversational context?

Granted I'm a bit of a Randall Munroe content addict, but it's become second nature now.

You know, that makes sense for a corporate network. They have an extremely aggressive firewall on the academic campus, which is how it should be.

However, they have failed to provide isolated networks for the research labs which just need it for even downloading LLMs (they have banned huggingface!).

Moreover, a hostel is residential. They should provide either the option of getting an external connection (which I would happily do!) or provide a means of non-stupid internet which they aren't.

Exactly.

I'm from a cybersec and devops background, and the IT admin here is just an ancient family-appointed person with no idea of how stuff works and with a lot to gain from under the table corporate dealings.

This is a man who believes that 15 megabit is sufficient bandwidth for CompSci students in their hostels (not the college, mind you, the hostel specifically) and decided that banning games was a "hero move".

Vendor locked into Sophos and a custom third party provider, these people have zero idea about what they're doing. I've met them various times and had various discussions up and down the org chart - this is a man who thinks he should have full access to every student's browsing history in their own time and that all VPNs are the same (he doesn't know how VPNs work btw) and allow for evasion from their network policies.

It's all a bit cursed because he fear-mongers the upper echelons of the college administration by showing them made up logs saying "students are hacking the network" to justify this.

Yes.

I agree.

There's such an opportunity for people to actually explore ideas whose prototyping cost would have been too high with both time/money to not be worth it earlier.

And even outside that perspective, there's a lot of broken corpo software now. The indie hackers are fighting back. See Helium by imputnet, for example. Ghostty by the revered Mitchell Hashimoto is another example of something I daily and is relatively indie.

Corpo-slop seems to be enshittifying at an exponential rate due to decision paralysis and general management talent decay.

I shifted from Crush to Opencode this week because Crush doesn't seem to be evolving in its utility; having a plan mode, subagents etc seems to not be a thing they're working on at the mo.

I'd love to hear your insight though, because maybe I just configured things wrong haha

One's banned in my hostel because of a stupid sysadmin.

One isn't.

Either via a VPN or a tunnel.

Yeah, it's actually quite decent, but I still feel that most of those features are basic UX that really shouldn't be application/vendor tied.

I found Goose (by Block) - https://block.github.io/goose - much better in this regard. Granted it perhaps doesn't have the app tie ins that most other providers do, but I can kinda just ask it to perform tasks in a specific kind of folder and it does, using whatever provider I want.

I got the GLM coding plan earlier, and given its generous rate limits, I found using it to do tedious tasks (like folder organization, which is perhaps my greatest weakness; especially Downloads) was a true addition to my productivity.

I feel there's a lot of scope for applications to focus on task contexts (with/without code agents having a ton of files like AGENTS.md, CRUSH.md, CURSOR.md spammed around) in specific folder bounds with proper user sandboxing.

Ah yes, the good ol' HN hug o' death.