Areibman

Alex here—- one of the team members behind the project.

We built this for the Llama3 Cerebral Valley Hackathon. The idea was this: My ~/Downloads folder is extremely messy, and I wanted an agent to fix it for me. So we built one.

LlamaFS reads file contents and metadata to Ollama with LlamaIndex, Moondream, and Whisper to understand what it’s about. It renames the files according to a specified pattern and organizes similar files into directories. Also, LlamaFS doesn’t overwrite anything until you explicitly ask it to (no risk of deleting anything precious)

One cool thing we benchmarked with AgentOps was the speed. It goes pretty fast, ~500ms per file.

Some others have mentioned before, but there are actually several projects working on exactly this:

https://github.com/Arize-ai/openinference

https://github.com/traceloop/openllmetry

https://github.com/Scale3-Labs/langtrace

It was a question more for us than general public. I think YC is probably the best first check you can get since they invest in companies with nothing more than an idea.

We got rejected (2nd time applying) and ended up raising a substantial seed/pre-seed round at excellent terms a month later.

Is there a compelling reason to still apply?

To what degree do lawyers actually want to track their time accurately?

I used to work a timesheet intensive job (consulting). The quiet rule was log the minimum 45h/w if you like your engagement; log your actual hours if you hate it.

When my lawyer bills me 1 hour, I know it doesn’t actually take him 1 hour. That’s just his rate. This has been my general experience with small shops. Do bigger firms pay closer attention to detail?

We knew the time would come, but we built ChatOCR in a week. If we overthought the time horizon problem, we'd have 0 users and $0.

But also, GPT-4-Vision is multi-modal but does not specifically use OCR. Our tool is used mostly to extract text from documents and load it into context, and we still saw growth after OpenAI built this feature into ChatGPT.

We use a third party plugins manager called pluginlab.ai. It manages auth and subscriptions for plugins users by prompting them to sign in when they hit a paywall.

As a (former) ChatGPT plugins developer, our business has absolutely tanked due to GPTs. Discoverability is nonexistent because search is just buried by spam. At least with plugins, there were only a few hundred to sort through, and most had some unique API they would plug into.

For context, we built ChatOCR- an OCR tool that lets users extract text from photos and PDFs. We made roughly $20k from 39,000 users over 6 months on the plugins catalog

Makes sense. Coframe (https://coframe.ai) comes top of mind

Very slick, straightforward demo. Question on my mind: The AI A/B testing space seems to be getting crowded. In what way do you see Just Words standing out?

I'm not sure I'd classify most of these as AI agents, it's mostly web crawlers. (Though the definition of agent is fuzzy enough that I suppose you could lump them in).

I'm also curious, will adding "Common Crawl" to the user-agent disallow list in your robots.txt actually do anything?

Added context: I run a tracker of vetted AI agents with verticalized use cases https://staf.ai

I just came back from Puerto Vallarta, and it's shocking how pharmacies advertise their drugs to tourists. I'm talking massive signs with every prescription med under the sun-- Adderall, Xanax, Retinol, you name it. All next to the HGH and testosterone posters featuring unreasonably muscular bodybuilders.

LLM observability strikes me as an extremely, extremely crowded space. And YC has funded an enormous number of them.

What do you think is the key differentiator between you and everyone else? Is vendor lock-in really that huge of an issue?

[0] https://hegel-ai.com, https://www.vellum.ai/, https://www.parea.ai, http://baserun.ai, https://www.trychatter.ai, https://talc.ai, https://github.com/BerriAI/bettertest, https://langfuse.com

Maybe I'm missing something here, isn't this largely achieved by ONNX already?

[0] https://onnx.ai

As others have echoed, VCs are usually looking for a unicorn exit scenario to offset the massive risk they're taking. You could consider other debt instruments such as convertible notes or even small business loans.

I've also heard good things about tineyseed https://tinyseed.com

Plenty more:

- https://www.getscorecard.ai

- https://arxiv.org/abs/2308.03688

- https://withmartian.com

- https://aihero.studio/

(author of agentops here)

Surely they were benchmarking on the credentials of the founding time moreso than the actual memo, right?

Baseten was by far the easiest setup I've tried https://www.baseten.co

I was following until the very end. This felt very handwavy:

>Now, since e is somewhere between 2 and 3, we have to conclude that n must be 2. Further, from our curve, we can see that there can only be one other value of x for which (lnx)/x = (ln2)/2, and since we know that 2⁴ = 4², we know that the other integer must be 4.

Is this sort of thinking something that could be justifiably written in a real proof? Sure, conveniently fishing out 2 as n seems to narrow down the search space for m, but I'd imagine this wouldn't work for harder problems

Wouldn’t it make more sense to shame people who pulled out of SVB while simultaneously advocating for a bailout?

Only a couple of prominent investors on that list fit the bill

Congrats on the launch!

Some thoughts:

- Options are severely limited. Only Next.js and MUI? If there's only 1 stack, doesn't feel like there's any AI in going on

- Would be nice to see some live template sites without having to connect my GitHub and run them manually.

- This feels like cookiecutter, but for multiple languages and a nice wrapper. The GPT bit doesn't interest me (well, from a technical standpoint it does)

- You're running up against the hundreds of template sites that already exist. What's the unique selling point? The configurability?

General question for the MLOps community: I usually see tools like this launch all the time. I'm eager to experiment, but I find my use cases never fit, and I end up just building a simple internal tool that does the job. Usually because the dataset or model is too unique.

Is it just me?

Related thread where the article runs chatGPT through a number of political alignment tests. (Didn't get as much attention though).

https://news.ycombinator.com/item?id=34038485

I think BlueJ is a fantastic intro to Java. Does a good job of teaching the concepts without creating the same IDE dependence of IntelliJ. And not as brutal as JGrasp thankfully.

Jupyter notebooks, surprisingly. I prefer the mix of markdown and LaTeX together. Way faster to just write and go.

I used to feel the exact same way until I saw what Circle did with USDC.

I’m highly doubtful that FedNow will be able to compete in terms of innovative new features. Say what you will about blockchain, there’s a lot of builders out there aggressively churning out new tech without anyone’s permission.

Does anyone sincerely believe FedNow will gain traction for consumer payments? The fees don’t seem particularly cheap (at least compared to something like Solana Pay), and the FedNow website does not inspire confidence.

Also somewhat interesting is the design of Ottoman coins. Whereas the Byzantines imprinted religious figures and rulers on the coins, Ottoman coinage completely lacked human depictions (not even of Sultans) and was imprinted with Ottoman Turkish script. Very similar to how Christian frescoes were plastered over in the Hagia Sophia

Replace `reddit.com` with `teddit.net`. Pages load normally and without ads

Gas taxes and tolls are in place to fund roads. (Though in many states it just ends up in the general fund).