serjester
serjester on Hacker News
It seems strange to write a blog post about being proven wrong, without a single example of being wrong. Especially with all the current Claude.md drama.
Everyone loves being wrong in the abstract - this reads more like a self-congratulation than a genuine reflection.
Exciting but it’s priced at 2.5X Sol - we haven’t seen pricing this high since GPT 4.5. We will see if the real world use cases outweigh the sticker shock.
Cursor is a victim of its own success. Clearly the future is agents but the average user just wants them for a smarter tab completion.
I also cancelled my subscription since I’m not particularly sold that I need a middle man between me and OpenAI / Anthropic.
Akur8 | AI Engineer | NYC (Hybrid) | Full-time
Akur8 builds the OS for insurance pricing - our SaaS platform helps 320+ insurers worldwide model risk up to 10x faster using transparent ML.
We're hiring an AI Engineer to join our team. We're building systems that transform tens of millions of messy, unstructured regulatory documents into actionable intelligence. If evals, agentic fine-tuning, and agent orchestration are the problems that light you up - not just things you've read about but things you actually have opinions on - we'd love to talk.
The team is small and talent-dense. You'll ship to production, talk to real customers, and have genuine ownership over what you build.
Come join us at https://jobs.ashbyhq.com/akur8. Mention in your application that you saw our post on HN.
There's a potentially amazing use case here around parsing PDFs to markdown. It seems like a task with insane volume requirements, low budget, and the kind of thing that doesn't benefit much from autoregression. Would be very curious if your team has explored this.
The vast majority of people don't need smarter models and aren't willing to pay for a subscription. There's an argument to be made that ads on free users will subsidize the power users that demand frontier intelligence - done well this could increase OpenAI's revenue by an order of magnitude.
This is going to be tough to compete against - Anthropic would need to go stratospheric with their (low margin) enterprise revenue.
This is a hand wavy article that dismisses away VLMs without acknowledging the real world performance everyone is seeing. I think it’d be far more useful if you published an eval.
Congrats on the launch - you're value-add is quite confusing as someone that's at the applied AI layer. This comes off as more of a research project than a business. You're going to need an incredibly compelling sales pitch for me to send my data to an unknown vendor to fix a problem that might be obviated by the next model release (or just stronger evals with prompt engineering). Best of luck.
Seems like their critique boils down to two areas - pandas limitations and fewer built ins to lean on.
Personally I've found polars has solved most of the "ugly" problems that I had with pandas. It's way faster, has an ergonomic API, seamless pandas interop and amazing support for custom extensions. We have to keep in mind Pandas is almost 20 years old now.
I will agree that Shiny is an amazing package, but I would argue it's less important now that LLMs will write most of your code.
If cloudflare goes down, you can blame them. If your hand rolled solution fails when cloudflare exists, you’re going to have a tough pitch to leadership why you’re in charge of the technical roadmap. Choose your battles, and this is not a hill worth dying on.
It's disappointing there's no flash / lite version - this is where Google has excelled up to this point.
As someone that also worked at a large automakers, I think you’re making large, unfounded assumptions.
Warranty data flows up from the technicians - good luck getting any auto technician to properly tag data. Their job is to fix a specific customer’s problem, not identify systematic issues.
There’s a million things that make the data inherently messy. For example, a technician might replace 5 parts before they finally identify the root cause.
Therefore, you need some sort of department to sit between millions of raw claims and engineering. I would be curious what kind of alternative you have in mind?
Seems like a very natural fit for fine tuning - would have loved to see more on the LLM side.
Did you try fine tuning the LLMs?
This is just a company advertisement, not even one that’s well done. They didn’t benchmark any of the real leaders in the space (reducto, extend, etc) and left Gemini out of the first two tests, presumably because it was the best performer (while also being multiple orders of magnitude cheaper).
A lot of the comments on here are too pessimistic. Deepnote has had the single best jupyter interface for years now - unfortunately locked behind a cloud subscription though. Jupyter itself has been stagnant for far too long, and it's much appreciated there's more options coming online that have a modern level of polish.
Marimo is great, but it's good to have competition in the space (especially when both projects are still owned and maintained by VC backed companies).
One of the big ones that hasn't been mentioned is all of a hexagon's neighbors are equidistant. As a result, h3 is a better fit for flow modeling - stuff like telematics. This has some nice properties for ML too.
You can see one of my jupyter notebooks that dives deep into this with h3 here: https://drive.google.com/file/d/18jIVEbE_1QbwTbHdMqj0AVqguf2...
H3o is an awesome piece of work. I created polars bindings for it(another reason I love polars) and last time I benchmarked it, it had 5X better performance than even duckdb’s C++ implementation.
https://github.com/Filimoa/polars-h3
I feel like interviews with Sam always leave you underwhelmed - OpenAI needs more compute, he’s excited AI will accelerate science, and some sort of subtle hint he doesn’t care about copyright holders, while saying the exact opposite.
Would have loved to see a deeper explanation of how they will make something cost competitive with the Nvidia / AMD and where Broadcom is helping.
This is the most confusing pricing page I’ve ever seen - different options have different credit usage and different cost per credits? How many degrees of freedom do you real need to represent API cost.
Congrats on the launch. YC 2022? I'm assuming this was a pivot - what lead to it and how do you guys plan on making money long term?
Always appreciated Taleb's quote on this:
> A good book gets better at the second reading. A great book at the third. Any book not worth rereading isn't worth reading
The big labs have 50+% margins on serving the models, the training is where they lose money. But every new model boosts OpenAI's revenue growth which is unheard of at their size (300+% YoY). Therefore it's completely reasonable to keep doubling down and making bigger bets.
Most people miss that they have almost a billion free users that are waiting to be monetized. Google makes 400B a year and it's crazy to think OpenAI can't achieve some percentage of that. Why would you slow down and let Google catch up for the sake of short term profitability.
Desperate pivot aside, I don't see how anyone competes with the big labs on coding agents. They can serve the models at a fraction of the API cost, can trivially add post training to fill gaps and have way deeper enterprise penetration.
Congrats on the launch guys, but is this any different from setting up usage based pricing in stripe? This kind of seems like a solution in search of a problem.
It’s absolutely insane they managed to raise 50M dollars. Now they’re shooting for a desperate AI pivot that no one wants.
It seems like the company has stagnated and Cook seems to consistently makes questionable decisions - fighting the app store rulings, doubling down on vision pro, focusing on OS redesigns instead of generative AI.
Maybe they've just calcified - the youngest person on their board is 63. At the end of the day they haven't had a credible threat in over a decade.