ZeroCool2u

Much more interesting link: https://deepmind.google/technologies/gemini/

Signal is one of the non-profits I happily donate to. Myself, my family, and my friends use it almost exclusively.

The title is misleading. They've dropped the proposal as applied to Chrome, but are still pursuing it for the Android WebView API, which is basically a wrapper around Chrome.

It seems odd that you can't even preview a small cached subset (10 rows?) of a table in Snowflake. I don't even remember a time I couldn't see a table preview in BigQuery, so surely it's not a technical limitation?

Is there no explanation as to why they removed support and added it as a MS Store bundle? Seems really strange.

I'd really like to see AV1 hardware accelerated encoding and decoding built into the M3 and see Apple standardize on it finally. The schism caused by H264 and H265 is just so annoyingly inconvenient.

They did! I used my personal first, which is just Firstname.Lastname@gmail.com which was the one that broke. The other one was Firstname@company.com which seemed to work fine.

The book is quite interesting, but it does seem like a lot of the underlying work is farmed out to GeoStats.jl, which doesn't really seem to use the same vocabulary I'd expect in other languages using PostGIS or Geopandas etc. For example, I don't see many mentions of Polygons or MultiPolygons when I search. However, I do find this page[1] which seems to define similar(?) equivalents. Can I expect equivalent geospatial joins/queries to be available? I don't see many mentions of the types that I would normally do, especially overlay operations[2].

[1]: https://juliaearth.github.io/GeoStatsDocs/stable/domains.htm...

[2]: https://geopandas.org/en/stable/docs/user_guide/set_operatio...

"Choose your model Requests for code generation are made via an HTTP request.

You can use the Hugging Face Inference API or your own HTTP endpoint, provided it adheres to the API specified here[1] or here[2]."

It's fairly easy to use your own model locally with the plugin. You can just use the one of the community developed inference servers, which are listed at the bottom of the page, but here's the links[3] to both[4].

[1]: https://huggingface.co/docs/api-inference/detailed_parameter...

[2]: https://huggingface.github.io/text-generation-inference/#/Te...

[3]: https://github.com/wangcx18/llm-vscode-inference-server

[4]: https://github.com/wangcx18/llm-vscode-inference-server

Just a heads up, I'm getting an error when I sign up. I'm using the Scitodate activation code that was emailed to me and it's claiming it's invalid.

Edit: Hmmm, worked when I signed up with a different email.

This is exactly how I use ChatGPT right now. Like an extreme rubber ducky for research when I don't have anyone else I can talk to about something. Also, thank you for the self promotion, I'll check out MirrorThink!

This is a good starting place https://marketplace.visualstudio.com/items?itemName=HuggingF...

This idea seems to track with the notion we saw early on in the use of ChatGPT where folks suggested that answers with more 'filler' language seemed to produce higher quality responses. The 'pause' token seems like a better way to do this, but it's still pretty bizarre how well it works.

Here's a gift URL: https://www.wsj.com/tech/meta-floats-charging-14-a-month-for...

There's an ONNX runtime, but to use the runtime you do need to convert your model into ONNX format first. You can't just run a TF of PyTorch model using the ONNX runtime directly. (At least last time I checked.) Unfortunately this conversion process can be a pain and there needs to be an equivalent operator in ONNX for each op in your TF/Torch execution graph.

Seems almost too good to be true, but I really hope it's not. How does it handle things like CUDA dependencies? Can it somehow make those portable too? Or is GPU acceleration not quite there yet?

Didn't this happen to Sony already once before?

"NVIDIA TensorRT-LLM is now available in early access and soon will be integrated into the NVIDIA NeMo framework—part of NVIDIA AI Enterprise, an enterprise-grade AI software platform with security, stability, manageability, and support. Developers and researchers will be able to access TensorRT-LLM through the NeMo framework on NGC or through the source repository on GitHub.

Note that you must be registered in the NVIDIA Developer Program to apply for the early access release. You must also be logged in using your organization’s email address. We cannot accept applications from accounts using Gmail, Yahoo, QQ, or other personal email accounts."

Okay, kind of a bummer.

I literally just bought this last night. Works with just a webcam and is shockingly accurate. https://beam.eyeware.tech/

Oh the cloud part is going to be a huge deal breaker in so many industries. This will automatically be blocked by default at my work place for certain.

Just my 2 cents and I generally agree with the sentiment in the comments about Postgres just being rock solid reliable and having great integrations / vendor neutral.

One of the huge benefits I see with Postgres is its extension system. PostGIS is amazing. Can save any company that needs to do heavy geospatial processing $$$ by dodging ESRI. The only database I've found that handles geospatial stuff better is BigQuery, but considering the revolution that was Google Maps over the best decade I don't think that would really surprise anyone.

BigQuery is ACID compliant and easily handles petabytes.

https://cloud.google.com/bigquery/docs/introduction

See: https://arxiv.org/abs/2306.15595

"We present Position Interpolation (PI) that extends the context window sizes of RoPE-based pretrained LLMs such as LLaMA models to up to 32768 with minimal fine-tuning (within 1000 steps), while demonstrating strong empirical results on various tasks that require long context, including passkey retrieval, language modeling, and long document summarization from LLaMA 7B to 65B. Meanwhile, the extended model by Position Interpolation preserve quality relatively well on tasks within its original context window. To achieve this goal, Position Interpolation linearly down-scales the input position indices to match the original context window size, rather than extrapolating beyond the trained context length which may lead to catastrophically high attention scores that completely ruin the self-attention mechanism. Our theoretical study shows that the upper bound of interpolation is at least ∼600× smaller than that of extrapolation, further demonstrating its stability. Models extended via Position Interpolation retain its original architecture and can reuse most pre-existing optimization and infrastructure."

From what I've read you can, but I haven't tried myself or looked into it too deeply.

At the cost of being very specific to Rust, Shuttle is pretty damn simple. https://www.shuttle.rs/

"We present Position Interpolation (PI) that extends the context window sizes of RoPE-based pretrained LLMs such as LLaMA models to up to 32768 with minimal fine-tuning (within 1000 steps), while demonstrating strong empirical results on various tasks that require long context, including passkey retrieval, language modeling, and long document summarization from LLaMA 7B to 65B. Meanwhile, the extended model by Position Interpolation preserve quality relatively well on tasks within its original context window. To achieve this goal, Position Interpolation linearly down-scales the input position indices to match the original context window size, rather than extrapolating beyond the trained context length which may lead to catastrophically high attention scores that completely ruin the self-attention mechanism. Our theoretical study shows that the upper bound of interpolation is at least ∼600× smaller than that of extrapolation, further demonstrating its stability. Models extended via Position Interpolation retain its original architecture and can reuse most pre-existing optimization and infrastructure."

This is a pretty wild result. Almost all of the open source models folks have been using from HF etc can suddenly easily be extended to have an enormous context window for little to no effort. You can already see these models being distributed in the most popular formats for inference, like GGML, on The Bloke's HuggingFace profile!

Should probably link directly to the announcement here: https://proton.me/blog/proton-pass-launch

I have already, but my broader point was that this seems like a really common use case that could help with adoption and growth.

This is a really cool idea and something I'd love to use, especially on large discord servers. Something that happened to me just yesterday was a friend of mine and I were trying to play a newly released game together and he was having a difficult to debug issue. The game is extremely popular and the dev team is literally 3 or 4 people. The main support channel is just a discord server, but that's incredibly hard to search for information about bugs and how people fix them.

This is exactly the kind of thing I'd love to see added to the server, so that finding the information in the massive chat channels I needed was easier and could be distributed to others.

If any Struct folks are around, the game is called BattleBit. The devs might be happy to hear from you!

I also saw this and was initially both equally horrified and impressed.