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

The writing is on the wall for handwriting recognition

newsletter.dancohen.org

Post reply on HN
Print view View original

111–116 of 116 posts

Re: The writing is on the wall for handwriting recognition

#111

Post by robotresearcher »

throwup238 wrote: post #73
ecshafer wrote: post #68

Earlier quoted context omitted.

If there is 1 job at a university. And there are 10 researchers applying. And 1 took this improvement in research speed to do more research, and 9 took the change to play more piano and take more walks, then most likely that one will get the job. This competitive nature is what has driven society forward and not kept us at just above subsistence agriculture.

> This competitive nature is what has driven society forward and not kept us at just above subsistence agriculture. The UN estimates that around 500 million households or 2 billion people are still subsistence farmers. In 2025. Fat lot of good competition has done them, especially when they don’t have enough surplus to participate in a market economy to begin with.

Their children are far more likely to survive childhood than at any time in history.

Re: The writing is on the wall for handwriting recognition

#112

Post by myth_drannon »

coredog64 wrote: post #54
myth_drannon wrote: post #45

Earlier quoted context omitted.

Right, it can do modern writing but anything older than a century ( church records and census)and it produces garbage. Yandex Archives figured that out and have CER in a single digit but they have the resources to collect immense data for training. I'm slowly building a dataset for finetuning TROCR model and the best it can do is CER 18% ... which is sort of readable.

How do you do, fellow TrOCR fine-tuner? I'm using TrOCR because it's a smaller model that I can fine tune on a consumer card, but the age of the model and resources certainly make it a challenge. The official notebook for fine tuning hasn't been updated in years and has several errors due to the march of progress in the primary packages.

I think I based my notebook on the official example but yes at some point new versions of the libraries completely broke it. I had to pin the versions for it to work again.

This one works, you can check the versions https://pastebin.com/QPjGHN8j

Re: The writing is on the wall for handwriting recognition

#113

Post by themaninthedark »

spiritplumber wrote: post #85
volemo wrote: post #61

Earlier quoted context omitted.

Better “thinking” computers will breed worse thinking people, huh?

Dr. Walter Gibbs: Won't that be grand? Computers and the programs will start thinking and the people will stop.

Look at GPS and then "self-driving" cars.

With GPS we have seen people confidently drive past road closed signs and around barriers off bridges.

With self-driving technology, we have seen them defeat safe guards so they can sit in the back while the car accelerates up to 70 in a subdivision.

Re: The writing is on the wall for handwriting recognition

#114

Post by themaninthedark »

jgeada wrote: post #98
BJones12 wrote: post #94

Earlier quoted context omitted.

Between 1965 and 1995 the average American gained about 6 hours per week of leisure time. They then used most of the additional free time to watch TV.

And what's happened since 1995 (30 years ago!) ? Because all the trends seem to indicate that to make a living people are working longer hours, holding multiple concurrent jobs (eg https://gameofjobs.org/are-americans-now-more-likely-than-ev... ), and holding off retirement.

We started offshoring manufacturing and growing the service economy?

Now the service economy is turning into the sharing economy, I think the only thing we are sharing is the greater profits and they are taking the lions share.

Re: The writing is on the wall for handwriting recognition

#115

Post by volemo »

f3b5 wrote: post #88
volemo wrote: post #61

Earlier quoted context omitted.

Better “thinking” computers will breed worse thinking people, huh?

Socrates allegedly was opposed to writing since he felt that it would make people lazy, reducing their ability to memorize things. If it wouldn't be for his disciple Plato who wrote down his words, none of his philosophy would have survived. So I'm not completely disagreeing with you, but I also am not too pessimistic, either. We will adapt, and benefit through the adoption of AI, even though some things will probabl…

> We will adapt, and benefit through the adoption of AI, even though some things will probably be lost, too.

“What doesn’t kill you, makes you stronger”. We will adapt and benefit, or we will not — time will tell.

Re: The writing is on the wall for handwriting recognition

#116

Post by red75prime »

akoboldfrying wrote: post #106
red75prime wrote: post #81

Earlier quoted context omitted.

Ask them to mark low confidence words.

Do they actually have access to that info "in-band"? I would guess not. OTOH it should be straightforward for the LLM program to report this -- someone else commented that you can do this when running your own LLM locally, but I guess commercial providers have incentives not to make this info available.

Naturally, their "confidence" is represented as activations in layers close to output, so they might be able to use it. Research ([0], [1], [2], [3]) shows that results of prompting LLMs to express their confidence correlate with their accuracy. The models tend to be overconfident, but in my anecdotal experience the latest models are passably good at judging their own confidence.

[0] https://ieeexplore.ieee.org/abstract/document/10832237

[1] https://arxiv.org/abs/2412.14737

[2] https://arxiv.org/abs/2509.25532

[3] https://arxiv.org/abs/2510.10913

Post reply on HN