Hacker Newsnew | past | comments | ask | show | jobs | submit | zzleeper's commentslogin

I run an internal OCR benchmark for historical research (old texts.. newspapers, typewriter, but also handwritten)

Google is BY FAR at the top of the Pareto frontier, perhaps b/c of Google Books.

But Sol 6.2 (low) is the first non-Google model that grazes that frontier. Sure, it's still more expensive than Hydrogen, but not too far! (and that's at the current low Hydrogen prices)


only DOJ/Trump can release them. Maybe they'll burn em


Congress can subpoena people.


Trump would give all of them executive priviledge


The Biden DoJ could have prosecuted, if there were evidence. Same set of files, I reason.


Maybe, but Merrick Garland was a worthless, toothless AG who couldn’t even prosecute the insurrection that happened on live TV. So maybe not.


He did the best job anyone could have with the election interference and stolen documents cases (slam dunk imo). I'm not a lawyer but it seems the defense strategy ran out the clock. What would you have proposed?

Garland would have launched the Epstein case if there were proof. It is possible an investigation started and did not reach the threshold required, we'll never know. If they cannot prosecute, the DoJ does not publicly attack people.


They could run out the clock because Merrick Garland spent too much time running his strategy of starting from the bottom and prosecuting regular foot soldiers. What he should have done was appoint Jack Smith or someone like him as soon as he was sworn in, rather than playing coy about trying to figure out how something like J6 happened and who was responsible.

It wasn't a time for following the facts "where ever they may lead", as he loved to say. The facts as they stood when Garland was sworn in already showed it was Trump who was responsible -- the impeachment record had collected enough evidence against Trump at that point to warrant a SC on day one of the Biden admin. Instead Garland waited almost half his term!

It became clear Garland really dropped the ball when the Congressional J6 Committee's investigation actually outpaced the Justice Department's investigation, which is a pretty sad indictment of that DOJ when Congress can get its shit together but the DOJ can't.


I think neither of us is a lawyer. If the groundwork wasn't laid properly for Jack Smith, the whole investigation could have been thrown out. I would assume Garland knew the optimal strategy.


They thought they were gonna win the election and wanted to wait until afterwards to prosecute trump. Didn't pan out


Please elaborate. There were two DoJ cases against Trump (election interference and stolen nuclear documents) plus several others from other states. Why would the Biden DoJ sit on an Epstein case until Biden 2?


That's actually a pretty good reason. If you are going to put him away for life anyway*, maybe you can hold off on the one which would drag victims into court.

*An inevitable outcome, slowed by the immunity ruling and Aileen Canon, that could only be permanently derailed by the unthinkable scenario where voters re-elect him.


I disagree that protecting the victims would have been a reason to delay prosecution. Garland is criticized for publicly launching the two prosecutions too late in Biden's term. He knows more about the law than I do so I'm sure there was a good reason.

It will be interesting to see if the prosecution resumes after Trump leaves office and if the immunity ruling will get him off the hook, then: "This case would have determined whether then-president Donald Trump and others engaged in election interference during the 2020 election, including events during the January 6, 2021, attack on the U.S. Capitol."


Biden was very concerned about losing faith in democracy. Putting Trump in jail before an election isn't a good look. Like a politician jailing his opponents.

Is that really that hard to understand??


That would imply the DoJ would also sit on the stolen nuclear documents and interference cases until Biden 2. I assume either case could have resulted in jail time.


And they could have arrested and tried Trump for treason after Jan 6, but decided against it because they are feckless "moderates" who seek bipartisanship even with the scum of the Earth. Doesn't mean anything about the amount of evidence (or lack thereof) against Trump in the files.


The tried him for election interference. Unfortunately the defense ran out the clock.

Has anyone this high profile been prosecuted for treason in modern times? They could not have kept him in jail pre-trial for four years for a complex treason trial.


Not in the US, but Jair Bolsonaro was.


I'm fine with using AI for code, where the quality will likely be better than those of academics.

But writing.. FFS everything I see is just AI slop. And I only realize when I start wondering why the paper is going and going and saying nothing, and then pangram tells me that everything past the intro is just AI. :/


The glass half full take is that we are very lucky that there is some much literature that pre-dates LLMs. Our brains are tuned to enjoy that more. People who grow up with LLM slop from the start won't necessarily see it the same way.


Yup. I can read novels from the 1st century [1] or from the 19th century [2] or from the 1970s [3] for the rest of my life (well, until they get banned).

[1] https://en.wikipedia.org/wiki/Callirhoe_(novel)

[2] https://en.wikipedia.org/wiki/Jane_Eyre

[3] https://en.wikipedia.org/wiki/Swords_and_Deviltry


Out of curiosity, which harness do you use? Hermes, ..?


I used private torrents for the best part of 10 years but life got in the way, and when I tried to go back after my last Pvt torrent closed (Morethan..) I realized the bar was much higher than before. I only managed to get into MAM and perhaps and some point that would open more doors and so on, but it's anything but trivial lol (unless you never lost access to Pvt torrents or maybe if your friends do)


MaM actually has one of the higher bars, but they have the most regular signups and the community to support it. That’s a good point of reference from 10 years ago, though.


A bit tired of spending $200 out-of-pocket for openai. What do you use as harness? (for me the harness if half of the benefit... controlling my PC, working from phone, etc.)


https://omp.sh/ has amazing defaults and it sips tokens. Works really well with DeepSeek V4.1 Flash.

Use the model through a fast and reliable provider such as Fireworks directly, skip OpenRouter.


Did you ever try pi by itself? For those new to the pi ecosystem - any rationale to go with pi vs omp?


It's like choosing between vim and helix. I started my career with vim in the early 2000's, customized the whole thing and had my config in a version control.

Then I installed helix and I just use it without config.

If you like configuring things take pi, if not omp is pretty much great defaults.


I use Claude Code + eternal terminal + tailscale + tmux + some custom skills to get notifications through nfty.sh.

I get the same UX on every platform, works perfectly on very low bandwith environments such as in a cabin, in the subway or in the middle of nowhere.

I tried using other harness such as Pi and opencode but I did not like them. If Claude Code gets weird I can swap in an instant.

You just need to follow this guide and disable artifacts in Claude Code's config: https://api-docs.deepseek.com/quick_start/agent_integrations...


What makes it run well on low bandwidth environments? Is it eternal terminal? (I am not familiar with that).

I found that ssh is pretty bad in these scenarios, so my usual herdr over a phone doesn't always work properly.

Now I'm using paseo which in principle solves my issues properly, but unfortunately it's "reconnection" and state sync is pretty slow (probably going over their servers).


Yes Eternal Terminal has been a godsend for bad connectivity and switching networks.

The other part is that since SSH only sends the changes on the screen, it uses very little bandwidth.

Try Eternal Terminal it's very easy to setup and is just a simple layer over SSH.


My main gripe is latency. Typing out letters with bad connectivity is very error prone - a native text box is all I would need, but I will give it a try.


I’m curious about your use of Tailscale, is that for you to reach a local LLM remotely from anywhere?


I have a big beefy desktop at home which I ssh (using EternalTerminal instead of raw port 22) into. I built it last summer right before the prices got very expensive. It's headless so I use a cheap Macbook Air to connect to it at home and I use Termux on my phone to continue working from everywhere.


Dude prompt your agent to set up always on remote connections via systemd… then you can drive from Claude or codex mobile apps natively over their native hookup. Works great.


I don't want to rely on Claude of Codex mobile features at all. Also none of this works when you use third party LLMs which is what the current comment tree is about.


This. There's so many better ways to drive a fleet of agents this way than using "remote" features of which I don't trust, anyway.


I have a server living in my home office, always on. I have a tmux session on it with vanilla Codex and Claude Code CLI, I can via my Ubiquiti network stack wiregaurd in to this box anywhere on the globe with just my laptop. Works super well for me. I also have some cheap shelley power plugs that I can use to cycle my PC’s power state if needed.


This is basically my setup but I'm using tailscale and zellij. I don't have any contingency plan in place for my power or home internet going down though..


I used to do the same but being unable to copy/paste easily, follow links, upload files, etc led me to write a small bridge so I can turn Github issues into my harness frontend:

https://github.com/skorokithakis/symphony

Now I chat with OpenCode/Pi over Github issues, which is a miles better experience.


OpenCode works nicely for me. You can connect it to the DeepSeek platform with an API key.


I use pi.dev, it is pretty minimal, but you can extend it however you want since agent has access to it's own documentation.


I use crush (https://github.com/charmbracelet/crush) with the following patches:

  curl https://tg.st/u/0001-fix-unblock-all-commands-in-bash-tool.patch | git am
  curl https://tg.st/u/0002-feat-add-light-theme-with-auto-detection-for-white-b.patch | git am
  curl https://tg.st/u/0003-feat-enable-yolo-mode-by-default.patch | git am
  curl https://tg.st/u/0004-fix-disable-mouse-grabbing-to-restore-native-termina.patch | git am
  curl https://tg.st/u/0005-feat-skip-project-init-prompt-and-quit-immediately-o.patch | git am
  curl https://tg.st/u/0006-feat-remove-scrambled-rune-animation-from-waiting-sp.patch | git am
  curl https://tg.st/u/0007-feat-remove-quit-banner-and-thank-you-message.patch | git am
  curl https://tg.st/u/0008-feat-show-output-in-full-instead-of-collapsing-trunc.patch | git am
  curl https://tg.st/u/0009-fix-discover-map-model-features-advertised-by-v1-mod.patch | git am
  curl https://tg.st/u/0010-feat-keep-large-and-small-model-selections-in-sync.patch | git am


Far left? Most economists consider it center-right, and it has been like that for more than 100 years.


I thought that was a joke.. Like how I used "delve into" back in 2024 or so lol


It might be a joke, but I'm sure there are lots of people who don't perceive it as such. For me, it pulled me out of the article and diluted its advice.


Honestly, it's a bit of a disappointment

- Many more mediocre papers written (mediocre ideas, implementation, claude-isms everywhere)

- Much easier to try every possible combination of a regression in order to show the result you want (same for theorists).

The one thing I'm happy about is it's now much easier to extract historical data from old documents from Google Books. Still not perfect, but takes you 95% there. And creating plots and datavis just for quick exploration is super fast.


(cross posted from the other announcment thread)

A big problem I have with OpenAI's models (and of course Claude) is that they tend to write the most over-engineered pieces of code, beyond the imagination of any architecture's astronaut.

Just this week I asked 5.6-sol-ultra to update a 1000 LOC python script I had, to "incorporate the key lessons learned when using it for another project".

I left it overnight and went to sleep. In the morning I realized it had created a monstruosity of 180 PYTHON SCRIPTS, with maybe 100,000 lines of code, each more crazy than the other. It took me minutes even to track where a single action took place, due to all the crazy imports, defensive coding, and premature optimization.

Similarly, anything they write is riddled with jargon that almost feel like they want me to give up trying to understand. Made up phrases that ended up with me having no idea of what was going on.

So now to my assessment: The reason why " Nobody Has Actually Built a Software Factory" [1], and why even SOTA LLMs struggle so much with open-ended unsupervised tasks is precisely this. They somehow let complexity explode, and unless it's also accompanied with an explosion in e.g. the number of agents, the amount of processing time, etc. then projects become broken/unmanageable.

Sure, LLMs are great at producing code that can be thrown out, so they are amazing when searching for exploits, for instance. But as of 5.6 they still lack either a better harness that encourages KISS principles, or a better RL step.

(And not sure why, but doubt Astra will fix this.. they seem to be aiming for AGI and for beating crazy benchmarks, which is not very aligned with KISS)

[1] https://news.ycombinator.com/item?id=49510843


Please don't copy-paste comments. It makes merging threads a pain, since we have to figure out what the duplicates are and then make sure the comment in the 'winning' thread gets all the replies.


noted, thanks dang!


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: