HNHacker News
TopNewBestAskShowJobs

hubraumhugo

9,168 karma · joined June 23, 2019

Building web datasets for humans & agents @ kadoa.com

Also made HN Wrapped (hn-wrapped.kadoa.com)

Happy to chat about data, AI, startups, or anything else: adrian@kadoa.com

submissionscomments
hubraumhugo··on On Writing Well
As an engineer who became a founder, I cannot recommend the book enough. Whether you're improving your landing page, writing your pitch deck, reaching out to customers, or developing your company and product strategy, communicating effectively in writing is a very crucial skill.

Invest in improving your writing skills. It will pay dividends in every aspect of your business.

hubraumhugo··on Fine-tuning now available for GPT-4o
Curious to hear some real-world use cases for fine-tuning and what your setup looks like in terms of:

- frameworks

- hardware

- foundation models (OS vs 3rd party)

- evaluation & training data

Did the performance gains justify the investment for fine-tuning?

hubraumhugo··on Launch HN: Sorcerer (YC S24) – Weather balloons that collect more data
Very cool to see more YC hard tech startups emerging [0]. These are the kind of moonshot projects I love to see getting funded (instead of just SaaS and AI).

[0] https://www.ycombinator.com/companies/industry/hard-tech

hubraumhugo··on Launch HN: Trellis (YC W24) – AI-powered workflows for unstructured data
You mention validation and schema guarantees as key features for high accuracy. Are you using an LLM-as-a-judge combined with traditional checks for this?
hubraumhugo··on OTranscribe: A free and open tool for transcribing audio interviews
Fascinating how traditionally very complex and hard ML problems are slowly becomming commodities with AI:

- transcription

- machine translation

- OCR

- image recognition

hubraumhugo··on Show HN: The First Non-Smart AI Pendant (NotFriend)
For those who don't get it: this is a parody of a recently launched AI wearable called friend https://www.friend.com
hubraumhugo··on OpenAI Announces SearchGPT
Too bad it's only a waitlist. iFixit apparently published fake repair guides [0] that then got crawled by ChatGPT for training [1]

[0] https://www.ifixit.com/Guide/Data+Connector++Replacement/147...

[1] https://chatgpt.com/share/e52dc4dd-77e6-48a5-a7ca-77e3dfa39e...

hubraumhugo··on Open source AI is the path forward
The big winners of this: devs and AI startups

- No more vendor lock-in

- Instead of just wrapping proprietary API endpoints, developers can now integrate AI deeply into their products in a very cost-effective and performant way

- Price race to the bottom with near-instant LLM responses at very low prices are on the horizon

As a founder, it feels like a very exciting time to build a startup as your product automatically becomes better, cheaper, and more scalable with every major AI advancement. This leads to a powerful flywheel effect: https://www.kadoa.com/blog/ai-flywheel

hubraumhugo··on Llama 3.1
I wrote about this when llama-3 came out, and this launch confirms it:

Meta's goal from the start was to target OpenAI and the other proprietary model players with a "scorched earth" approach by releasing powerful open models to disrupt the competitive landscape.

Meta can likely outspend any other AI lab on compute and talent:

- OpenAI makes an estimated revenue of $2B and is likely unprofitable. Meta generated a revenue of $134B and profits of $39B in 2023.

- Meta's compute resources likely outrank OpenAI by now.

- Open source likely attracts better talent and researchers.

- One possible outcome could be the acquisition of OpenAI by Microsoft to catch up with Meta.

The big winners of this: devs and AI product startups

hubraumhugo··on Joe Biden stands down as Democratic candidate
Hard to draw a line here in terms of guidelines:

> Off-Topic: Most stories about politics, or crime, or sports, or celebrities, unless they're evidence of some interesting new phenomenon.

https://news.ycombinator.com/newsguidelines.html

hubraumhugo··on Use a work journal
It's surprising how many devs are trying to overoptimize for productivity with various fancy tools and techniques, when it actually comes down to simple basics.

Here is what I rely on as a founder who does a lot of context switching:

- a never-ending text file for todos and work journal [0]

- calendar for planning (and blocking focus time)

- website blocker

- turned off notifications

[0] https://news.ycombinator.com/item?id=39432876

hubraumhugo··on General Electric HTRE-3 Nuclear Jet Engine
Sitting here with my morning coffee and reading about a nuclear jet engine I've never heard of makes me realize how special HN is. Please, never change!
hubraumhugo··on My finetuned models beat OpenAI's GPT-4
> Data extraction/structuring data is the only serious application of LLMs

I fully agree. I realized this early on when experimenting with GPT-3 for web data extraction. After posting the first prototype on Reddit and HN, we started seeing a lot of demand for automating rule-based web scraping stacks (lots of maintenance, hard to scale). This eventually led to the creation of our startup (https://kadoa.com) focused on automating this "boring and hard" problem.

It comes down to such relatively unexciting use cases where AI adds the most value.

AI won't eliminate our jobs, but it will automate tedious, repetitive work such as web scraping, form filling, and data entry.

hubraumhugo··on Show HN: R2R V2 – A open source RAG engine with prod features
Do you also see the ingestion process as the key challenge for many RAG systems to avoid "garbage in, garbage out"? How does R2R handle accurate data extraction for complex and diverse document types?

We have a customer who has hundreds of thousands of unstructured and diverse PDFs (containing tables, forms, checkmarks, images, etc.), and they need to accurately convert these PDFs into markdown for RAG usage.

Traditional OCR approaches fall short in many of these cases, so we've started using a combined multimodal LLM + OCR approach that has led to promising accuracy and consistency at scale (ping me if you want to give this a try). The RAG system itself is not a big pain point for them, but the accurate and efficient extraction and structuring of the data is.

hubraumhugo··on Car dealerships revert to pens and paper after cyberattacks on software provider
My entrepreneurial mind immediately thinks: Let's create a better solution to revolutionize this industry and make the old, expensive incumbents obsolete. We can drive to the affected car dealers and migrate them one by one.

Of course, I'm not going to do it and it's probably extremely hard :)

hubraumhugo··on Training AI
It's been almost three years since the release of GPT-3, and there still isn't any regulation or standardization for opting out/in of AI training.

Can we expect any changes in this area at all? Can anyone point me to some resources?

hubraumhugo··on MLow: Meta's low bitrate audio codec
Is it just my perception or has Meta become cool again by sharing a ton of research and open source (or open weights) work?

Facebook's reputation was at the bottom, but now it seems like they made up for it.

hubraumhugo··on Show HN: Revideo – Create Videos with Code
Just curious, are you the founders of https://haven.run (YC S23)? I noticed that the Linkedin company page now redirects to Revideo.

Would you mind sharing a bit about your pivot? I always find these stories interesting!

hubraumhugo··on Designing a Lego orrery
Whenever I see a project like this that involves hundreds of hours of engineering and craftsmanship, I think: Doing nothing after retirement doesn't appeal to me, so this is how I'd love to spend my time (even sooner if circumstances allow).
hubraumhugo··on Show HN: I created this land use visualisation for the Netherlands
A treemap would probably be the better choice. The Swiss government uses one to visualize their land use: https://www.bfs.admin.ch/bfs/en/home/statistics/territory-en...
hubraumhugo··on ThankYouHN: 14 Years
My own startup wouldn't exist if I hadn't made a HN post with a crappy MVP and received feedback from users, founders, and potential customers.

It's a very special place that successfully avoided its "Eternal September". I hope dang knows how much we appreciate his work to not let the quality go downhill.

hubraumhugo··on If A.I. Can Do Your Job, Maybe It Can Also Replace Your CEO
I'm a bit surprised that we currently see so many sensational headlines on HN. There was just another one on the front page: "Anthropic Chief of Staff: These next 3 years might be the last few that I work."

AI is useful and not just hype, but we don't see any larger shifts in the labor market or economy.

Will AI automate tedious repetitive tasks, such as form filling, data analysis, and data entry? Yes, absolutely.

Will AI take your job? Unlikely, at least in the near future.

hubraumhugo··on What we've learned from a year of building with LLMs
I think it comes down to relatively unexciting use cases that have a high business impact (process automation, RPA, data analysis), not fancy chatbots or generative art.

For example, we focused on the boring and hard task of web data extraction.

Traditional web scraping is labor-intensive, error-prone, and requires constant updates to handle website changes. It's repetitive and tedious, but couldn't be automated due to the high data diversity and many edge cases. This required a combination of rule-based tools, developers, and constant maintenance.

We're now using LLMs to generate web scrapers and data transformation steps on the fly that adapt to website changes, automating the full process end-to-end.

hubraumhugo··on ChatGPT Edu
I remember the fascination of reading through physical encyclopedias when I was a kid.

Having the collective knowledge of the world in my pocket and a personal tutor for any topic still sounds surreal.

hubraumhugo··on What We Learned from a Year of Building with LLMs
Agree, that's a very good summary. Would love to see some benchmarks for the two approaches.
hubraumhugo··on What We Learned from a Year of Building with LLMs
Comprehensive and practical write-up that aligns with most of my experiences.

One controversial point that has led to discussions in my team is this:

> A common anti-pattern/code smell in software is the “God Object,” where we have a single class or function that does everything. The same applies to prompts too.

In theory, a monolithic agent/prompt with infinite context size, a large toolset, and perfect attention would be ideal.

Multi-agent systems will always be less effective and more error-prone than monolithic systems on a given problem because of less context of the overall problem. Individual agents work best when they have entirely different functionalities.

I wrote down my thoughts about agent architectures here: https://www.kadoa.com/blog/ai-agents-hype-vs-reality

hubraumhugo··on Hate Chatbots? You Aren't the Only One
This article could just as well be from 2016[0], when Facebook first launched chatbots for their Messenger. I played around with NLP tools like Dialogflow back then to build one of these.

We've come a long way since, but some (UX and tech) challenges remain the same.

[0] https://techcrunch.com/2016/05/29/why-do-chatbots-suck/?gucc...

hubraumhugo··on How to enhance generative AI's problem-solving capabilities, boost productivity
I recently wrote a blog post about why "AI agents" are still too early, too expensive, too unreliable: https://www.kadoa.com/blog/ai-agents-hype-vs-reality

The WebArena leaderboard[0], which benchmarks LLM agents against real-world tasks, shows that even the best-performing models have a success rate of only 35.8%.

[0] https://docs.google.com/spreadsheets/d/1M801lEpBbKSNwP-vDBkC...

hubraumhugo··on Cloudflare took down our website
The Cloudflare CEO & co-founder is quite active on Twitter[0] and somewhat active on HN[1], would be interesting to get his perspective on this.

[0] https://x.com/eastdakota/

[1] https://news.ycombinator.com/user?id=eastdakota

hubraumhugo··on Show HN: I made a online free tool to enhance and auto-crop your screenshots
You're doing exactly the right thing to kickstart your career as a dev/founder/indie maker: Building a small useful tool and launching it (that's the important part, many never get to this stage), seeking feedback, and improving based on it. This is the essence of the "build, measure, learn" methodology that every startup tries to apply.

In fact, my own startup wouldn't exist if I hadn't made a HN post with a crappy MVP and received feedback from founders, investors, and potential customers.

← PreviousPage 4 of 14Next →