castles

Location: Cambodia (UTC+7)

Remote: Yes

Willing to relocate: Potentially

Technologies: Python/TypeScript of late, C/C++, C#, Java, Scala, JS, Kotlin, Android, More

Résumé/CV: LinkedIn (https://www.linkedin.com/in/niqd/details/experience/), GitHub (github.com/sublimator)

Email: ndudfield@gmail.com

Curious debugger, doer, and hands-on maker of things, creative and self-reliant.

Ideal for early-stage companies. Worked remotely since 2011 from Cambodia for U.S. companies, including SublimeHQ/Ripple/Coil and smaller NFC/RFID startups. Easily managed 12hr timezone differences, and open to working within EU time zones for even better alignment.

Like many, I am presently interested in AI(esp. browser++), however I remain open to anything interesting.

Thanks.

Super Kind folk, and Brave too. I once offered to build something for free for them to prove I could build things, but they said there was no need and kindly paid me. I didn't get the gig in the end, but in this cold age of frauds, I really appreciated that.

"YET ANOTHER shiny new toy to distract me. Can't help myself even though I think it's mostly a waste of time"

Am I just projecting? Relatable, in any case :)

  Location: Cambodia (UTC+7)
  Remote: Yes
  Willing to relocate: Potentially
  Technologies: TypeScript of late, C/C++, C#, Java, Scala, JS, Kotlin, Android, Python, More
  Résumé/CV: LinkedIn (linkedin.com/in/nicholas-dudfield-3b938629), GitHub (github.com/sublimator)
  Email: ndudfield@gmail.com

Curious debugger, doer, and hands-on maker of things, creative and self-reliant. Ideal for early-stage companies. Worked remotely since 2011 from Cambodia for U.S. companies, including SublimeHQ/Ripple/Coil and smaller NFC/RFID startups. Easily managed 12hr timezone differences, and open to working within EU time zones for even better alignment.

Like many, I am presently interested in AI(esp. browser++), however I remain open to anything interesting.

Thanks.

thanks, I'll give it a crack

It seems "local" is all you need :)

Interesting! What's the prompt eval processing speed like compared to llama.cpp and kin?

nexusraven/starling-lm/chat.lmsys.org, all from UC Berkeley and affiliates

"Chatbot Arena is an open-source research project developed by members from LMSYS and UC Berkeley SkyLab"

Something in the water there?

OpenAI also had years "flying under the radar" training on API requests

Segment ids are all you need (probably)

> accepting/rejecting tokens etc.

That sounds interesting

I'm still working on RAG (been annoyingly ill last while) but you might find this interesting to keep an eye on: https://socontextual.com/

Ollama /JUST/ recently added some (fast) embeddings models: https://ollama.com/library/nomic-embed-text

Oh, I saw this on The Bloke discord Nifty!

> figuring out what a market segment might need

What segment do you imagine being the most likely to use this? What existing alternative processes? How does this improve upon them?

> my now lawyer would lament about the power of an os-interface to the hash function.

What feedback did your lawyer have on this. Will this help in legally binding claims?

> available on the Dibs website the same as if the transaction was created with our interface

Can anyone take a file and check on the site to see who has logged it? The video showed that url and other optional contact metadata could be attached

> Learning how to summarise effectively might save hours per week and improve decisions about which sources deserve our limited time/attention.

If you like summaries, you'll probably love de-summaries (WIP): https://socontextual.com/

Wow, sounds like a fun/cool project. Wish you great success!

To clarify - did you mean Mixtral (8x)7b, or Mistral 7b?

Just don't work like a founder and it could be good.

Working with friends you can be totally honest and comfortable with can be amazing.

More context would help. Your age? Startup stage? You have kids? Etc

I had similar ideas on the back burner for per site scripts. MV3 sure makes things annoying at times! I want to build something that's kind of a mix between what you're doing here and what I've been experimenting with.

Did you experiment with the sidePanel api yet? You can open it, but the <input> will not get focus like with the popups. You'd think at least with the commands keybindings, or reacting to the browser action icon click this should be possible?!

I asked Lumos some questions and it's hitting the embeddings endpoint many times, and then when I asked a follow up question, it hit the same endpoints again, each taking a few hundred ms. It's not caching them? TBD?

p.s. iirc There's some way to set the key in the manifest such that you'll get a static id, useful for OLLAMA_ORIGINS where you don't need * in the examples.

p.p.s starling-lm is really great for a default local model for these types of things

edit: https://github.com/andrewnguonly/Lumos/pull/38

I've been working on something in the same vein, though just using manual selections, or selections auto expanding from the visible viewport.

I really like the per-site parsers, which make a lot of sense

Congrats :)

By default it will summarise the context, but you can give it arbitrary queries, and it will respond with references to the source. It's handy if you've developed a healthy distrust.

More platforms would be a bit of a drain. Just one lonely dev on this. Multiple LLMs was indulgent and distracting enough.

Next? There's tonnes you could do, but finding something that won't leave you with a massive bill, and or offend people's sense of privacy is a bit tricky. For the latter it was "designed" so you explicitly select what text to give the AI. This can get annoying, so there's auto selection commands to select up to a model's context limit.

- Relax on the beach

- Non contiguous contexts - even "simply" expanding your selection from the viewport and down can be quite complicated due to things like fixed headers - it's hard to find a single selection that spans all the on-screen elements

- Replacing the window.prompt box

- Follow up chats to the original query

- Fine tune models for the text zoom so it works much faster (without needing a lot of hidden "thoughts" ) and more accurately

Tough?

The line mapping was much more annoying to work out than you'd probably imagine

If it were a weaker country? You can imagine the response might be different?

All you need is an importance to note

Are they wrong :) ?

It's not super obvious, but even with Plus you can opt-out of training.

Aside: If you can see other colleagues' interactions with the custom/private GPTs, it could be quite an efficient way to share knowledge, especially for people in disparate time zones.

almost certainly - edit: https://github.com/ggerganov/llama.cpp/tree/master/examples/...

Have seen too many instances of "too many chiefs" to instantly dismiss the idea. Perhaps voting on a time limited dictator/final arbiter?