Cursor if I recall actually started life as a VScode plugin. But the plugin API didn’t allow for the type of integration & experiences they wanted. Hit limits quickly and then decided to make a fork.

Not to mention that VSCode has been creating many "experiemental" APIs that are not formalized for years which become de facto private APIs that only first party extensions have access to.

Good thing that Copilot is not the dominant tool people use these days, which proves that (in some cases) if your product is good enough, you can still win an unfair competition with Microsoft.

Yeah, that’s very well put. They don’t store black-and-white they store billions of grays. This is why tool use for research and grounding has been so transformative.

Definitely, and hence the reason that structuring requests/responses and providing examples for smaller atomic units of work seem to have quite a significant effect on the accuracy of the output (not factuality, but more accurate to the patterns that were emphasized in the preceding prompt).

I just wish we could more efficiently ”prime” a pre-defined latent context window instead of hoping for cache hits.

Confident idiot: I’m exploring using LLM for diagram creation.

I’ve found after about 3 prompts to edit an image with Gemini, it will respond randomly with an entirely new image. Another quirk is it will respond “here’s the image with those edits” with no edits made. It’s like a toaster that will catch on fire every eighth or ninth time.

I am not sure how to mitigate this behavior. I think maybe an LLM as a judge step with vision to evaluate the output before passing it on to the poor user.

I had a similar result trying to create 16 similarly styled images. After half a dozen it just started kicking out the same image over and over again no matter what the prompt said. Even the “thinking” looked right, but the image was just a repeat. I don’t know if this is some type of context limitation or what.

I got around it by using a new prompt/context for each image. This required some rethinking about how to make them match. What I did was create a sprite sheet with the first prompt and then only replaced (edited) the second prompt.

I still got some consistency problems because there were a few important details left out of my sprite sheet. Next time I think I’ll create those individually and then attach them as context for additional prompts.

Oh smart. This is good guidance. Yeah fascinating how longer running context causes these side effects, especially the repeated image with no changes bug.

Whats your thoughts on the diagram as code movement? I'd prefer to have an LLM utilize those as it can atleast drive some determinism through it rather than deal with the slippery layer that is prompt control for visual LLMs.

I think that's the right approach and what I've been experimenting with. Diagram as code and then style transfer from output diagram to desired look. That's where I've had the most success.

Have you considered that perhaps such things simply are not within its capabilities?

I mean, one of its flagship features is to make precise edits to images. And it's really good at it... until it randomly isn't.

Yes, same here.

I don't know if it's a fault with the model or just a bug in the Gemini app.

same. i gave it a very well hand drawn floor plan but never seems to be able to create a formal version of it. Its very very simple too.

makes hilarious mistakes like putting toilet right in the middle of living room.

I dont get all the hype. am i stupid.

When async notifications arrive from background processes… How is the user notified? (Not defending toasts, just curious how to do it better.)

GitHub seems to suggest banners or “Also consider ways to notify the user in other communication channels such as email, notifications, or a push notification in the GitHub app.”

On MacOS… emails and push notifications create… toast messages

Wait I don’t get it. CF is free for your use case, but you are looking for a cheap CDN? What’s cheaper than free?

I’ve got quite a few (very old) downloads coming to my websites, and I don’t want people to lose them. I also want to maintain a no-frills, store-and-forget thing that does not cost much. I was using AWS S3 + CloudFront for a long time, but I realized I was paying over $10 a month for something I didn’t even interact with or check often enough.

I’m OK with a sub-$10/mo budget, but Amazon Web Services doesn’t offer a way to pay recurring charges with Indian Cards. After this thread, I read up a lot yesterday and realized that Amazon AWS India is now pretty well oiled and working. I might stick to it and pay it off in advance. I’d be more than glad to, say, pay off $100 a year and not think about it.

The cost on AWS, I realize, is the S3 storage. Cloudflare is already fronting the CDN aspects of it.

So, I was looking for something with a sweet spot, say, pay something in the lines of $10 a year, at max about $25 a year, and it just dumps all of my files now and, to an extent, in the long term.

They are obviously looking for something to meet there future needs, not there current needs. It is free now, but they might be planning to one day tick over the free thresholds.

Everything in tech becomes ad-supported bullshit even if you pay for it. Tech knows no other business model with consumers besides devaluing the product to grow large enough to be another shitty ad platform.

I thought it was kinda funny that Google Slide’s own built in “beautify this slide” button converts the whole slide into an uneditable image.

AFAIK -- even the "Designer" feature of Microsoft Powerpoint (now folded under Copilot license I believe) gives slide deigns with shapes etc that are not editable. Thankfully the text remains editable. But if we wnat to ever so slightly modify the suggested design my removing or reshaping some if the shapes ... nopes. Feels like they are worried about humans with taste ripping-off the AI output :D

In some ways, the fact that the technology will shift is the problem as model behavior keeps changing. It's rather maddening unstable ground to build on. Really hard to gauge the impact to customer experience from a new model.

For a JS dev, it’s just another Tuesday

Is JS dev really still so mercurial as it was 5 to 10 years ago? I'm not so sure. Back then, there would be a new topic daily about some new JS framework etc etc.

I still occasionally see a blip of activity but I can't say it's anything like what we witnessed in the past.

Though I will agree that gen AI trends feel reminiscent of that period of JS dev history.

I’m working on a couple apps using Typescript and for me (ex-JS hacker coming back to it after some years) it’s still an insane menu of bad choices and new “better” frameworks, some of which are abandoned before you get done reading the docs. Though I get that it probably moved faster a few years ago.

I settled on what seemed like the most “standard” set of things (marketable skills blabla) and every week I read an article about how that stack is dead, and everybody supposedly uses FancyStack now.

Adding insult to injury, I have relearned the fine art of inline styles. I assume table layouts are next.

To lurch back on topic: I’m doing this for AI-related stuff and yes, the AI pace of change is much worse, but they sure do make a nice feedback loop.

If it is, it’s entirely self inflicted today. There’s some tentpole tech that is reliable enough to stick with and get things done. Has been for a while.

We can't help but overreact to every new adjustment on the leader boards. I don't think we're quite used to products in other industries gaining and losing advantage so quickly.