theolivenbaum

That's great! I'll need to investigate how to integrate this to https://tinder4cats.com/ now

Nice! Last time I found something like this was a 15$/mo yearly subscription, which was clearly targeted at real state agents, and didn't make any sense for a one time check... Luckily they just believed the system clock, so at least I could check over the year by changing the computer time

If you're looking for something local, we develop an app for macOS and Windows that let's you search and talk to local files and data from cloud apps: https://curiosity.ai For the AI features, you can use OpenAI or local models (the app uses llama.cpp in the background, it ships with llama3 and a few other models, and we're soon going to let you use any .gguf model)

Seems like a lot of it came from them inadvertently posting some internal API to GitHub: https://github.com/googleapis/elixir-google-api/commit/078b4...

Let me guess, Apple is having to pay OpenAI for this, and they want Google to pay them for it instead

Interesting article buried in the comments there about the fraud https://www.econcrises.org/2016/11/29/cendant-corporation/

Also very useful for ships, including commercial transporters

They probably also know that some types of queries are not ad-worth (like age of an actor, etc), while others would never show a box

We've been using a library called TNT for this job and are quite happy with the flow: https://github.com/pragmatrix/tnt

You basically annotate all your strings with a "my string".t() or t($"my interpolated string: {var}") and then use their CLI to extract the strings to be translated. It even includes google translation API support for you to kickstart the process.

Mono-repo, mono-language, and really enjoying it - enables much easier refactoring and using a single strongly typed language makes lots of issues go away. We do have a few other repos for open source projects or libraries we publish, but the majority of the work is done on our mono-repo.

Awesome work! You should put it behind a CDN like Cloudflare and see what's the cache hit ratio

Reminds me of a post I saw sometime ago: https://bendauphinee.com/writing/2023/12/30/to-know-where-yo...

We experimented with it sometime ago for our https://curiosity.ai app, initial training on your data was a bit heavy (at the time, probably fine by today's standards) but nice results if you had enough files. Needs to be done with care as for small datasets as there's not enough info for a model to learn and you end up introducing more noise than anything.

What I find interesting from the analysis is that the hard-coded, per-app event handlers suggests their integrations are nothing more than traditional wrappers around the services they support, instead of some magic, AI powered LAM living on the cloud that can interact with anything:

UberEventHandler SpotifyEventHandler FoodDeliveryEventHandler MidjourneyEventHandler TranslatorEventHandler SearchWeatherInfo SearchStockInfo

What kind of latency do you observe when using Gemini with such large contexts?

We're using in a large engineering use-case, 100s of millions of objects - and it almost doubled the nDCG@10 score vs pure vector search

Sold out in the EU already? Any ideas when it's going to be available again?

Interesting to see special tokens for tool calling:

"Function calling and special tokens This tokenizer includes more special tokens, related to function calling :

[TOOL_CALLS] [AVAILABLE_TOOLS] [/AVAILABLE_TOOLS] [TOOL_RESULT] [/TOOL_RESULTS]"

Is this something new on this model or has it been done already for a while?

*under $1m annual revenue and <1m users

That looks expensive compared to what groq was offering: https://wow.groq.com/

> I'm in search of a local LLM that can run completely offline for processing personal documents. Key requirements include privacy (no data leaves my machine) and performance (efficient with large datasets). Any recommendations for open-source / commercial solutions that fit the bill in 2024? Also, what's the current state of local LLMs—are: Are they practical and useful, or still facing significant limitations?

We've added support for it in our app if you wanna give it a try: https://curiosity.ai

And obviously the Germans have a word and a law for it: Erbpacht (https://de.m.wikipedia.org/wiki/Erbpacht)

We built a RAG system for one of our clients in the aviation industry. >20m technical support messages and associated answers / documentation, and we're seeing between 60-80% recall for top 3 documents when testing. Definitely pays off to use as much of the structure you'll find in the data, plus combining multiple strategies (knowledge graph for structured data, text embeddings across data types, filtering and boosting based on experts experience, etc). The baseline pure RAG-only approach was under 25% recall.

Hit a funny on Edge the other day: every time a page would open the browser "don't close this tab" dialog (like if you need to save something), the browser window would go on a crazy loop of resizing 1px width and repositioning itself 1px to the left, till it arrived on the left border of the monitor.

Hetzner just added GPU servers, might be worth checking too

You can try https://curiosity.ai, supports Windows and macOS

We just added local LLM support to our curiosity.ai app too - if anyone wants to try we're looking for feedback there!

Google ads is indeed a mess - I was surprised to see this is their cash cow, as the experience is just horrible.

Got access the other day to their latest AI models, and while everyone in the market will let you a) create an API token and b) use it, for Google you've a mess of service accounts, badly named permissions, services that change name all the time, 'just works' authorization that's basically a lock in to their own cloud. After spending a few hours with their docs, I just gave up - not worth the effort for now.

We're now also being asked to revalidate our integrations with them in the name of security (no change on our end, thousands of users using it every day). Another absolutely useless, bureaucratic, Kafkaesque process, with rules changing all the time, and not properly applied.

Google has long lost its way on software...

We support both in our app and enterprise product. The APIs (OpenAI) vs libraries (i.e. llama.cpp for on-device) are so similar that the switch is basically transparent to the user. We're adding support for other platforms APIs soon, and everything we've looked so far is as easy to integrate as OpenAI - except Google that for some reason complicates everything on Google Cloud.

Same here - incidentally was also one of the weirdest interactions with customer support I've ever had. I suspect the first point of contact was some sort of LLM/chatbot that desperately wanted to make sure I was feeling fine and that there was nothing to worry about. When I was forwarded to the billing support team the interaction went back to normal - couple of messages back and forth and some homework to set the real budget limit (the quota is just for alarms) and they waved the charge.