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

I just asked GPT6-Astra to transcribe a page from Newton's Principia to LaTeX. It did an amazing job outputing a XeLaTeX doc with a mixture of text and TikZ. Looks really good.

Did you do a pixel diff against the original? The output could look plausible at a glance but contain hard-to-spot errors or inaccuracies.

Unchecked OCR can lead to amusing results: https://news.ycombinator.com/item?id=18374359

Nice! Where is it? Thanks for sharing. (Also hopefully not in GitHub , it’s always down)


Me too. Claude Opus 5's English is insufferable. Opus 4.6-4.8 was more reasonable.

I've moved to Codex 5.6-Sol. Much saner English, much better at execution, and gets stuff done in a matter-of-factly kind of way (Claude Code is a mess these days -- it gets things wrong and goes around in circles).

But I'm harness agnostic and am not locked in. I just keep my issues in Kata Tracker (https://www.katatracker.com/) and switch harnesses/model when I need to.

Being loyal to a particular model/harness seem unwise to me.


> Me too. Claude Opus 5's English is insufferable. Opus 4.6-4.8 was more reasonable.

I suspect this degradation is happening because the AI labs are using the LLM's output to feedback into the input, to create a thinking loop, and they're optimizing that.


> The population of the west who fit into zoomer near millennials wouldn't even be close to 500m.

There are many people of all ages not in the west from very populous nations who use Insta, FB and WhatsApp.

I’ve learned that US usage patterns are not representative of rest of world.

More saliently, HN users’ conception of others’ usage patterns are absolutely not representative.


I lived in Montreal as a poor student and would eat ground horse because it was affordable (Quebec has a horse meat industry - viande de cheval). It tasted like a leaner ground beef and cost the same.

I also ate canned snails which were also affordable (CAD1.99 per can if I remember correctly).


I do this in one word “eli5” (explain like I’m 5). It’s a Redditism that it understands.

I also have a writing steering file that makes Opus’ writing less insufferable. Otherwise it’s really bad.

I also have an interlocutor skill that makes it less epistemically arrogant (ie Less Wrong asshole tendencies). With this skill I can have a real discussion with it instead of it trying to one up me.


Believe it or not I've had success with "ELI12".-


Same. I once asked for ELI5 and it rewrote the report using dinosaurs and construction vehicles...

I regularly find it helpful to say things like "Write with ELI10 clear sentences and established technical terminology (e.g. API)" or sometimes just "Write it like you are explaining to a colleague" works well.


I'm one of those Canadians with a Ph.D. who left for the U.S.

I didn't actually want to leave, but I was staring at the prospect of not unemployment, but underemployment.

Canadian industry doesn't really that many companies that can absorb Ph.Ds. It's not a money issue -- I was willing to accept Canadian pay. This was ca 2010s, and there were very few homegrown Canadian companies that had enough scale to need high level R&D (this was before Shopify and the rest). It was either you did your own startup and found your own funding or you worked entry level engineering jobs doing things like testing or audit work.

All the design and original engineering work happened at HQ in the US or Europe -- the Canadian office is usually just a branch office.

In the US, it's analogous to Chicago offices of big tech firms. They're just marketing and sales offices -- the real engineering work happened on the coasts (SF/Seattle or NYC).

So I reluctantly left Canada for a US company that although not famous, had a global footprint. Whatever I worked on immediatley had global impact because engineering was done in the US office and replicated worldwide. I learned a lot about how to operate at scale and over international markets.

Canada is a small country that has an education system that punches above its weight -- Canadian schools are excellent. But it has an industry that unfortunately doesn't. Somehow Canada has tried and failed a few times, with Nortel, with Blackberry, etc. to develop champions. Something is preventing Canadian companies from scaling. Even far smaller countries like Sweden have companies that are global (IKEA, Saab, Volvo etc), so I'm not sure why we can't do it too.

Maybe the old saying that Canadians are "hewers of wood and drawers of water" has some truth -- it's easier to extract resources than to compete on differentiated product.


That is absolutely hilarious. Dario Amodei, a romcom lead.


Echoing the other folks, I have a different experience.

I profile sql performance and LLMs find more opportunities than I could. All it takes is real data, a sql repl and an agent. Just ask the agent to use the repl to EXPLAIN and profile the sql. It works amazingly most of the time.


Meal is a bit hard to define.

Here are some cases:

You can get food for under $10 at many casual NYC restaurants -- if you just order an appetizer.

Also a snack is not a meal. For instance, I noticed a Salted Egg Yolk Bun in NYC for $1.50. But that's not really a meal (unless you get 4 of them). https://cheapfoodmap.com/spot/golden-steamer-new-york-ny

The average check price in NYC Chinatown can easily be under $10 if you get the right combination of appetizers/snacks.

Furthermore there are places where you can get food by the pound. Technically you can get a prepared meal at Whole Foods ($12.99/lb salad bar) if you get it just under 0.76 lbs. If you're light eater, then every Whole Foods in the country qualifies as cheap eats. Then the map will be saturated.


Yes, this is #1 homework this week to solve. How do we define equal standard for all food types?


I just fed this to GPT 5.6 Sol:

  Counterexample to Jacobian conjecture:

  ((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)

GPT wrote some SymPy code to check it. The response?

"As written, this is an explicit counterexample to the Jacobian conjecture. I checked it using exact symbolic algebra.

I do not see an algebraic catch in what you typed. Unless a term or exponent differs from the intended expression, it appears to disprove the conjecture. This deserves serious independent checking rather than casual dismissal."

Waiting for someone to write the Lean proof.



...no need for any lean here


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

Search: