Completely free flow typing is risky in terms of interpretability, but type narrowing - var a : supertype; if (a is subtype) { // a is known to be subtype }, or type case, saves boilerplate in any OOP language.
Yup, tried all variations of that. There's the advantage that you can use a lower hallucination OCR specific model for the pre-processing, at least for clean text. But for something hard like handwritten forms, applying VLM with context is less error prone than preprocessing to text.
Also - and this is bizarre - the token cost of doing that is higher, not lower, at least in Gemini world, and by a large margin. That's very counterintuitive, but a page encoded as image tokens can be smaller than same page as text, and is not meaningfully lossy on documents that are just typed text because the models are well trained on those.
We sure did. It's a great writer, better in a harness, will process lots large context, but complex reasoning with convoluted rules and low hallucination tolerance? That's still larger model territory.
That's fine if you're doing interactive dev tasks, but we're in the large volume, cost effective, big inputs, business still with low error tolerance business, and tuned the heck out of what we can get with minimal fix cycles. Millions of cases at hundreds of thousands tokens each - after all the prefiltering by cheaper models - and the tasks still need them to do convoluted reasoning. So 'usually get it right the first time' is a big part of the cost equation.
seems to respect reduce motion just fine! the animation, while extremely lovely, also does seem to be quite taxing on my GPU (~60% usage on M1 Max with Zen/Firefox, on a 1440p widescreen, which is admittedly quite a lot of space to render animation)
love the website though, i hope i one day will be able to make something that elegant!
Every technical topic involves AI now. Even if it's about why the ancient Greeks didn't turn their steam engines to industrial use - the AI's should be researching the details. We might as well ask which topics involve humans or use language.
TreasuryDirect's login and account recovery experience has been notorious for years, both for user experience and for people easily getting locked out for weeks. It's good they're being careful with this rollout, as it serves both individual and institutional accounts where dollar amounts involved are epic even by bank standards, and rarely checked by hand, so even single account breaches are serious.
The VLM's are so good at complex document understanding now. But you just can't trust them not to invisibly censor sensitive clinical/legal docs, even at the maximally permissive settings.
And the deep learning OCR-only models won't censor, but can and do hallucinate. I've yet to see a 'scan with different approaches and reconcile and say you're not sure if they don't agree' system just work for generic complex documents.
The way my harness set it up is going through 2 or 3 providers, and cross-checking through them, also with plain text extracted if available.
I think we also had a layer that for any quote extracted tested it back if it exists within the original.
If you wanted 100% accuracy, I think it wouldn't be too difficult nowadays to guess the font&size&other text settings, and re render the crucial parts.
The only mainstream-ish language where that still happens? R. Too bad, there's a lot more use cases for this sort of thing now - versioning, anything non-persistent agents touch, collaboration, auditable enterprise LOB. It's 2026, why are we (or our agents) still writing serialization code? Even if the AI's write the boilerplate, the state management fragility is often a tax/risk.
3.5 Flash-Lite seems available in US region, as was 3.5 Flash; but 3.6 Flash looks Global only so far when pinging. If Google employees are watching, will this issue go away?
reply