You don't. When you see it saying it stored something "in memory" you can just ask it to document it in the product documentation somewhere, and in future please keep doing that. Of course the note to keep doing that needs to go somewhere, which is what "memory" is for.
I just can't fathom how a multi-billion dollar company like Cloudflare could forego their design standards. They seem to have updated their landing page [0], but it's still very sloppy.
My sister's boss fired her (small education) company's web designer.
When you do things like that, you can't escape the slope even when you realize your mistake. (Since it's an education company, people notice illiterate garbage).
tbh, they don't really need the website, it could just be a simple pure text website with a simple tagline explaining what they do, a sign in and sign up button, that's it.
Cloudflare is a complete disaster these days, its quite an embarrassing fall from grace. I feel for the few engineers that still remotely care about their craft that work there.
I'm actually loving it more and more, but the disconnect in docs is disconcerting. Try connecting a jurisdiction-bound bucket to their AI Search and can't read from it - and their docs suggest in some places it should be fine and in others it is not. Go to submit a bug report and it's extremely high cost load to get information, even a basic writeup, shared, requiring github account / public repo and more.
They can probably get away with it at this point. Cloudflare already has the name and reputation, so I doubt a sloppy landing page costs them much business.
Really? I'm more surprised when I see a large company still have design standards. It seems it's all the rage to fire all your UX people and have your devs wipe up a UI because "UI is so easy."
Follow design standards? Fuck no, we need to STAND OUT instead. If you find the design hard to use as a result of not following standards your just an old curmuddgen.
I've seen a couple companies already change their design standards to "best in class Tailwind designs from Claude, use this prompt file/Agent file/skill file combo for all UX work".
Looking at automated emails I get from Microsoft and GitHub it certainly feels like they are one weird layoff from doing it company-wide.
The worst part is these glassification Tailwind heavy slop almost universally fail WCAG guidelines (and personally, seem hard to read to me because of taking the whispy sans serif thing to its foregone conclusion by almost exclusively using what to me are unreadably thin weights of the already too thin Inter font). The slop is going so quickly backwards for companies that invested in accessibility and these companies need to be taken to task for forgetting design isn't just aesthetics but readability and accessibility.
I feel you, especially on the WCAG stuff. It's so damn basic and yet, they refuse to do it. The regressions on usability and consistency across the design space is just so maddening.
And honestly, I'm not even that mad at AI. The way these companies are going I'll take an AI that is taught to follow the most basic design guidelines over what most companies shovel out these days.
Like fine, you want to stand out? Use different colour schemes or unique fonts (but like one or at most two fonts). Change your corner radius from time to time to freshen up. There are so many ways to make your UI "different" but still consistent and following basic best practices.
Some of the stuff throw out now a days would fail UI 101 assignments.
Negative response to low-effort AI-involved work is not what “AI doomerism” means.
A college student who is also very clearly (fron the links beyond the blog) a very active and enthusiastic AI user that thinks common (if correctly identified as less than ideal) UI trends from before vibecoding was common are indicators of “AI slop” and not just design being the done by people who aren’t very good at it (which, to be fair, includes but is not limited to AI slop) is unsurprising and nothing like doomerism.
You can disagree with the factual interpretations ans subjective interpretation in the piece without trying to tie it into a completely unrelated social phenomenon.
The site is utter shit. Total amateur hour. Terrible contrast, barely anything guiding the eyes, the padding to the sides is inconsistent. So is the entire design language from leading with some numbering that never goes beyond 01, round buttons, round input elements, interrupted by somewhat decorated layered elements which then leads to some kind of terminal window that doesn't resemble any terminal UI I've ever seen or used. It's all filled by various sloppy mentions of keywords that barely form some coherent narrative.
This page looks like some junior cobbled it together based on a hastily written ticket by some overworked product guy.
Hard policy, light enforcement in practice means there is an actual shadow policy that you don't know, which is pretty bad to the point that it would be illegal if the government did it.
Curious about your comment on two counts (genuinely, not being argumentative):
1. how is this bad, in Codeberg's case, practically speaking?
2. in what cases is this illegal for governments to do? In practice no legislation is perfectly enforced, & even judiciaries have varying degrees of exactness across countries - to the point that things such as desuetude are densely studied legal topics.
(note: I'm aware of the negative ramifications if desuetude but not aware of any negative ramifications of Codeberg's specific ai policy)
>in what cases is this illegal for governments to do?
It's legal certainty principle and equality before the law principle and also prohibition of discrimination principle.
For the law to be lawful (weird, but it's a thing, right) it should satisfy certain criteria. Me, being subject to the law should be able to predict whether my actions are allowed or not when I plan to do or not do them. If I can't predict the outcome of my actions, the law isn't valid. One of the outcomes of it is being able to read the law -- it's published and it's comprehensible, it has advance notice, etc.
If the rule says "don't host LLM generated code", but everybody hosts LLM generated code and then I say mean things about one of the people providing the hosting and suddenly I'm deplatformed for hosting LLM generated code, then I'm not deplatformed for LLM generated code, I'm deplatformed for saying mean things about people with fragile ego. Or being part of a specific group. You are often explicitly prohibited to make certain decisions based on belonging to a certain group (religious, ethnic, sexual, trade union, health, criminal record based), but it could happen that somehow the outcome of the shadow policy correlates with those. Since we don't know the shadow policy and the stated policy doesn't seem to be applied equally, this is not great.
See example about weed in a different comment.
>how is this bad, in Codeberg's case, practically speaking?
They look silly for stating it and not following (loss of credibility), then once they suddenly start following it, there will be different interpretations, including "I'm banned because I'm $protected_thing_that_you_should_not_be_banned_for", which is bad press.
> suddenly I'm deplatformed for hosting LLM generated code, then I'm not deplatformed for LLM generated code, I'm deplatformed for saying mean things about people with fragile ego
This is ultimately a matter of framing though. You can equally argue that (a) that person shouldn't be deplatformed because others' LLM use is tolerated or (b) all of them should be deplatformed including that person. In the latter case (b) the person who said mean things still experiences the same outcome, but by your logic this is just. Ultimately whether the others get deplatformed doesn't directly affect them, so what's the real difference here?
> > how is this bad
> they look silly
> bad press
This seems highly subjective & ultimately not really bad by any metric that matters.
>Ultimately whether the others get deplatformed doesn't directly affect them, so what's the real difference here?
The difference is -- if I know the effective policy, I can plan better and choose a different platform in advance (if the it's option (b)).
>This seems highly subjective & ultimately not really bad by any metric that matters.
Subjective opinions matter because people act on them (more) consistently than on objective facts. I would for example act on a subjective opinion that a platform provider is a bunch of racist assholes or affiliated with r-country.
The difference is that without consistent enforcement, the rule becomes just a fig leaf for arbitrary decisions by the moderation team, who then receives an enormous amount of unintended power. Most reasonable people wouldn’t argue that the moderation team should be allowed to ban anyone that they have a personal disagreement with, for example. Yet here we are.
The problem is for the platform as a whole, not the individual who may or may not be punished.
Governments have a lot of laws that they don't enforce, or enforce only lightly. For example, weed is still illegal in Netherland, despite many decades of tolerating its use.
I've heard it is only illegal becausr they signed treaties that would require them to keep it illegal, and they would have to leave these treaties. The USA president at the time hated weed and put it in every treaty he could.
It's illegal but you can't be prosecuted for usage or possession (unless you do it in a blowverbod area in front of a village cop that is having a bad day) -- the judge will ask why this person specifically is prosecuted if everybody around is smoking and will throw it away. It already happened in fact.
Their contention seems mainly focused on the heavy resource use of AI tooling.
They run a service: they could impose limits on the number of submissions, builds, branches, scans, etc, and give additional usage credits to the large community projects they want to encourage.
That & the legal aspects of hosting code of questionable copyright.
The LLM revolution has proven that if you have enough institutional investors, the Berne Convention doesn't apply to you, but it's sofar unclear whether that exemption from international law extends to small EU-based non-profits.
But they don't want to. They want to provide all needs for approved projects. They want that if a FOSS project legitimately releases 50 times a day, they want to make it work. If a slop project is wasting those resources that could be used for KDE, on nonsense, then they don't.
They're not a free for all, they do KYCP (know your customer's projects) with an initial assumption of good faith.
Also they're a voluntary society you can join by contributing money. They're not a charity as such, more like a group of people pooling resources, and also offering them to people outside the group on a good faith basis.
Every single issue they brought up in their blog post starts with even just 1% AI code. If they actually believed the absolutely absurd stuff they spew, banning it all would be the only logical solution.
The current solution solves none of the issues, AND pisses off most sane people (mostly because their arguments are fabrications). They dropped the cake and shit on it too.
Jev has good use cases that require speed. But speed is not all. Tasks like code review need quality, which costs time. If you just want speed, you might as well go with a non-reasoning model like Haiku.
By choice. I currently work at Stripe (opinions my own, I'm just some guy) and was here when it released. It's a genuinely useful tool, which explains the widespread adoption.
Or OpenAI just buys them outright. Buying your upstart competitor seems to be in the Silicon Valley Ten Commandments. The Fed whussed-out on breaking up FB and Insta last year, so there's never going to be any kind of remediation to worry about.
And for Jev, everyone has a price, and OpenAI's raised an historical amount of funding.
My first impression is that it's not so good at following prompt directions. I asked it to place a 3D text made of glass in a particular city. It instead gave me a broken 3D text on a white background. Maybe with different seeds it gets better, but it's more of a trial and error process than reliable results.
You could try attaching other images as references (I think you can attach a maximum of 10 images). If the attachments can be blurred or sketchy or generic enough, they could be used for generalization.
This is exactly what language models are good at. It's not a problem that could be solved purely deterministically, because it requires knowledge of the actual pronunciation in that particular language or dialect. There can be so many exceptions to the rule, so it's just easier to train a small language model on a sufficiently large dataset.
Sure, but note that a tiny LLM is many order of magnitudes larger than algorithms like this. It might well work better, but you're probably paying for very, very marginal improvements with a solution that's thousands of times larger and slower, unless you're you're being really creative and using a fairly non-LLM network of perceptrons or other ML technique. But it sounds like a fun challenge; if you or anybody feels like exploring that, I bet it'd get quite a few curious clicks here!
It really depends on the use case, which the author didn't hint to (sure, they say procedurally generating text, but exactly how and for what reason?). But whatever it is, I assume it needs to be accurate to some degree, and having a deterministic ruleset is just too subjective and superficial to cover all cases.
It doesn't need to be a full-fledged LLM either. What I mean is that a probabilistic machine learning model trained on a sufficiently large dataset is probably a better choice here. Small language models are pretty efficient nowadays, even for on-device use cases with constrained resources.
Well, I did once scrape wikipedia and classify based off simply off string suffix (storing those where longer suffixes disagreed with the shorter rule), and that's quite simple and effective - https://eamonnerbonne.github.io/a-vs-an/AvsAnDemo/ - and since it's based off actual statistics, it correctly distinguishes textually subtle stuff like "a NASA scientist" vs. "an NSA analyst".
There are structural limitations (i.e. this implementation never looks at preceding context and sometimes that matters, nor does it understand other clues like punctuation or multi-word suffixes). Nevertheless, the accuracy is high enough that I'm not sure it'd be easy for a small model to beat it, especially not without considerably more work to make sure your inputs cover more context (which in principle the plain statistics approach could likely deal with too).
If the whole point of multi-layer networks is to deal with weirdly shaped, non-obvious manifolds in a very high-dimensional space, then this problem just doesn't look that difficult and perhaps does not need that mathematical finesse: just store the prototypical examples and you're pretty much there without anything fancier.
reply