Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A new Google model is nearly perfect on automated handwriting recognition

generativehistory.substack.com

Post reply on HN
Print view View original

121–130 of 328 posts

Re: A new Google model is nearly perfect on automated handwriting recognition

#121

Post by muixoozie »

conception wrote: post #68
muixoozie wrote: post #65

Earlier quoted context omitted.

Dunno if you're right, but I'd like to point out that I've been reading comments like these about every model since GPT 3. It's just starting to seem more likely to me to be a cognitive bias than not.

I haven’t noticed the effect of things getting worse after a release but definitely 2.5’s abilities got worse. Or perhaps they optimized for something else? But I haven’t noticed the usual “things got worse after release!” Except for when sonnet had a bug for a month and gpt5’s autorouter broke.

Yea I don't know. I didn't mean to sound accusatory. I might very well be wrong.

Re: A new Google model is nearly perfect on automated handwriting recognition

#122

Post by ceroxylon »

xx_ns wrote: post #45

Am I missing something here? Colonial merchant ledgers and 18th-century accounting practices have been extensively digitized and discussed in academic literature. The model has almost certainly seen examples where these calculations are broken down or explained. It could be interpolating from similar training examples rather than "reasoning."

The author claims that they tried to avoid that: "[. . .] we had to choose them carefully and experiment to ensure that these documents were not already in the LLM training data (full disclosure: we can’t know for sure, but we took every reasonable precaution)."

Re: A new Google model is nearly perfect on automated handwriting recognition

#123

Post by schiffern »

efitz wrote: post #12

I haven’t seen this new google model but now must try it out. I will say that other frontier models are starting to surprise me with their reasoning/understanding- I really have a hard time making (or believing) the argument that they are just predicting the next word. I’ve been using Claude Code heavily since April; Sonnet 4.5 frequently surprises me. Two days ago I told the AI to read all the documentation from my…

  >I really have a hard time making (or believing) the argument that they are just predicting the next word.
It's true, but by the same token our brain is "just" thresholding spike rates.

Re: A new Google model is nearly perfect on automated handwriting recognition

#124

Post by krackers »

shwaj wrote: post #51
orionsbelt wrote: post #47

Earlier quoted context omitted.

Ilya Sustkever was on a podcast, saying to imagine a mystery novel where at the end it says “and the killer is: (name)”. Saying it’s just a statistical model generating the next most likely word, how can it do that in this case if it doesn’t have some understanding of all the clues, etc. A specific name is not statistically likely to appear

Can current LLMs actually do that, though? What Ilya posed was a thought experiment: if it could do that, then we would say that it has understanding. But AFAIK that is beyond current capabilities.

Someone should try it and create a new "mysterybench". Find all mystery novels written after LLM training cutoff, and see how many models unravel the mystery

Re: A new Google model is nearly perfect on automated handwriting recognition

#125

Post by krackers »

viftodi wrote: post #120
throwup238 wrote: post #2

I really hope they have because I’ve also been experimenting with LLMs to automate searching through old archival handwritten documents. I’m interested in the Conquistadors and their extensive accounts of their expeditions, but holy cow reading 16th century handwritten Spanish and translating it at the same time is a nightmare, requiring a ton of expertise and inside field knowledge. It doesn’t help that they were of…

You are right to be skeptical. There are plenty of so called windows(or other) web 'os' clones. There were a couple of these posted on HN actually this very year. Here is one example I google dthat was also on HN : https://news.ycombinator.com/item?id=44088777 This is not an OS as in emulating a kernel in javascript or wasm, this is making a web app that looks like the desktop of an OS. I have seen plenty such projec…

Every time a model is about to be released, there are a bunch of these hype accounts that spin up. I don't know they get paid or they spring up organically to farm engagement. Last time there was such hype for a model was "strawberry" (o1) then gpt-5, and both turned out to be meaningful improvements but nowhere near the hype.

I don't doubt though that new models will be very good at frontend webdev. In fact this is explicitly one of the recent lmarena tasks so all the labs have probably been optimizing for it.

Re: A new Google model is nearly perfect on automated handwriting recognition

#126

Post by nestorD »

throwup238 wrote: post #31
nestorD wrote: post #5

Earlier quoted context omitted.

Oh! That's a nice use-case and not too far from stuff I have been playing with! (happily I do not have to deal with handwriting, just bad scans of older newspapers and texts) I can vouch for the fact that LLMs are great at searching in the original language, summarizing key points to let you know whether a document might be of interest, then providing you with a translation where you need one. The fun part has been b…

> The fun part has been build tools to turn Claude code and Codex CLI into capable research assistant for that type of projects. What does that look like? How well does it work? I ended up writing a research TUI with my own higher level orchestration (basically have the thing keep working in a loop until a budget has been reached) and document extraction.

I started with a UI that sounded like it was built along the same lines as yours, which had the advantage of letting me enforce a pipeline and exhaustivity of search (I don't want the 10 most promising documents, I want all of them).

But I realized I was not using it much because it was that big and inflexible (plus I keep wanting to stamp out all the bugs, which I do not have the time to do on a hobby project). So I ended up extracting it into MCPs (equipped to do full-text search and download OCR from the various databases I care about) and AGENTS.md files (defining pipelines, as well as patterns for both searching behavior and reporting of results). I also put together a sub-agent for translation (cutting away all tools besides reading and writing files, and giving it some document-specific contextual information).

That lets me use Claude Code and Codex CLI (which, anecdotally, I have found to be the better of the two for that kind of work; it seems to deal better with longer inputs produced by searches) as the driver, telling them what I am researching and maybe how I would structure the search, then letting them run in the background before checking their report and steering the search based on that.

It is not perfect (if a search surfaces 300 promising documents, it will not check all of them, and it often misunderstands things due to lacking further context), but I now find myself reaching for it regularly, and I polish out problems one at a time. The next goal is to add more data sources and to maybe unify things further.

Re: A new Google model is nearly perfect on automated handwriting recognition

#127

Post by oasisbob »

muixoozie wrote: post #117
colordrops wrote: post #75

Earlier quoted context omitted.

Why would you assume cognitive bias? Any evidence? These things are indeed very expensive to run, and are often run at a loss. Wouldn't quantization or other tuning be just as reasonable of an answer as cognitive bias? It's not like we are talking about reptilian aliens running the whitehouse.

I'm just pointing out a personal observation. Completely anecdotal. FWIW, I don't strongly believe this. I have at least noticed a selection bias (maybe) in myself too as recently as yesterday after GPT 5.1 was released. I asked codex to do a simple change (less than 50LOC) and it made a unrelated change, an early return statement, breaking a very simple state machine that goes from waiting -> evaluate -> done. Howev…

That sounds more like availability bias, not selection bias.

Re: A new Google model is nearly perfect on automated handwriting recognition

#128

Post by oasisbob »

conception wrote: post #14

I will note that 2.5 pro preview… march? Was maybe the best model I’ve used yet. The actual release model was… less. I suspect Google found the preview too expensive and optimized it down but it was interesting to see there was some hidden horsepower there. Google has always been poised to be the AI leader/winner - excited to see if this is fluff or the real deal or another preview that gets nerfed.

I noticed the degradation when Gemini stopped being a good research tool, and made me want to strangle it on a daily basis.

It's incredibly frustrating to have a model start to hallucinate sources and be incapable of revisiting its behavior.

Couldn't even understand that it was making up non-sensical RFC references.

Re: A new Google model is nearly perfect on automated handwriting recognition

#129

Post by blharr »

ceroxylon wrote: post #122
xx_ns wrote: post #45

Am I missing something here? Colonial merchant ledgers and 18th-century accounting practices have been extensively digitized and discussed in academic literature. The model has almost certainly seen examples where these calculations are broken down or explained. It could be interpolating from similar training examples rather than "reasoning."

The author claims that they tried to avoid that: "[. . .] we had to choose them carefully and experiment to ensure that these documents were not already in the LLM training data (full disclosure: we can’t know for sure, but we took every reasonable precaution)."

Even if that specific document wasn't in the training data, there could be many similar documents from others at the time.

Re: A new Google model is nearly perfect on automated handwriting recognition

#130

Post by HarHarVeryFunny »

CamperBob2 wrote: post #90
HarHarVeryFunny wrote: post #84

Earlier quoted context omitted.

> Predicting the next word requires understanding If we were talking about humans trying to predict next word, that would be true. There is no reason to suppose than an LLM is doing anything other than deep pattern prediction pursuant to, and no better than needed for, next word prediction.

How'd you do at the International Math Olympiad this year?

I hear the LLM was able to parrot fragments of the stuff it was trained to memorize, and did very well
Post reply on HN