jerrygenser

> So while I will deny their green card application and not submit it,

I might be missing this in the thread. What is the reason that you deny their green card application and not submit?

Have you considered using HNSW instead of brute force?

Another thing I've seen is that some of these old school apps are sending certain requests that don't modify the page but set server side context which subsequent requests are dependent on.

For example, set context to a particular group and then subsequent navigation depends is filtered on that group even though the filter is not explicit on the client side but due to state stored in the session remotely.

This can also have implications on concurrency for a given session where you need to either create separate sessions or make sure there is some lock on particular parts of server side state.

Would this type of this eventually be possible? Or at least hooks in able for us to add custom code such as session locks

I was talking about web apps. But yeah, for old school desktop apps or windows native proxy MITM works

Will this work for SSR applications? e.g. think old school net or jsp apps which make network requests then receive HTML which then needs to be parsed in order to understand the key pieces of information and then additional network requests?

I've found it relatively straight forward to reverse engineer SPA requests however with server side rendered apps, yow would your service handle that?

> The r1 device and rabbit OS will be officially unveiled on January 9, 2024 in an online launch event on the sidelines of the Consumer Electronics Show (CES 2024) in Las Vegas.

I think this page is a bit dated. It mentions January 9, 2024 in future tense.

I was initially surprised that Khosla would be giving more funding given their experience over the past year.

I was just looking into htmx the other day. Came across the following library: https://github.com/iwanalabs/django-htmx-components

It's an example using django-components. Does this satisfy your comment about component at all?

where managers visit the homes of workers out sick

This would be another example of open source. Not from such a large company but a good reference including code, data, weights, etc.

https://allenai.org/olmo

It's surprising that the JR engineer who presumably has access to chat jippity submitted this bad code to begin with. Wouldn't they have had the AI review the code first?

would there be a scenario where you prefer to mark it as capex? opex has the advantage of reducing tax immediately and getting the tax benefit right away. I feel like that's always an advantage

I can confirm an approach like this works in practice.

Although instead of levenshtein I use spellfix (maybe it uses that under the covers? not sure). If there is no match from the first search, I use the sqlite spellfix extension [0] to find matches. Then feed those candidates into the terms.

https://www.sqlite.org/spellfix1.html

Love gamersnexus humor. The coupon code on their video: THISISFINE

Olamma currently has only one "supported backend" which is llama.cpp. It enables downloading and running models on CPU. And might have more mature server.

This allows running models on GPU as well.

They reference parquet files, not sure if it's only CSV or CSV even figures in that heavily other than the first iteration before migrating to spark

Azure document AI accuracy I would categorize as high not "mid". Including hand writing. However for the $1.5/1000 pages, it doesn't include layout detection.

The $10/1000 pages model includes layout detection (headers, etc.) as well as key-value pairs and checkbox detection.

I have continued to do proofs of concept with Gemini and GPT, and in general any new multimodal model that comes out but have it is not on par with the checkbox detection of azure.

In fact the results from Gemini/GPT4 aren't even good enough to use as a teacher for distillation of a "small" multimodal model specializing in layout/checkbox.

I would like to also shout out surya OCR which is up and coming. It's source available and free for under a certain funding or revenue milestone - I think $5m. It doesn't have word level detection yet but it's one of the more promising non-hyper scaler/ heavy commercial OCR tools I'm aware of.

I have a UI app with a backend API server. Sometimes I want clients to update their UI to get new features (UI payload is cached in the browser)

I have a version in the UI and the backend. If the UI is on an older version then I can either force reload the client or display a prompt to the user saying there's a new version of the app.

Is it more complex with react server components?

You're better off trying to figure out what features you like about articles that are less ambiguous as signal. You would then be able to finetune models to classify whether those features are present. Whether it's classificaton of chunks/sentences/tokens. For these, a bert model could be fine tuned to efficiently detect it.

They are 100% better for classification at a given compute budget. They can account for information before and after e.g. a token for token classification and use that information to classify.

Yes but you can use an llm to label data and then train a bert model which then costs a small fraction of time and money to run than the original llm.

Yes but Google hasn't needed gpus to power search. It's used cpu even for ML inference and training before deep learning.

He's going to make backends pluggable. So other more scalable backends than sqlite could work

> In terms of monetization, the goal is to just have a hosted queue system. I believe this can be cheaper than SQS without sacrificing performance. Just as Backblaze and Minio have had success competing in the S3 space, I wanted to take a crack at queues.

are you monetizing this as a separate business from: https://www.ycombinator.com/companies/scratch-data

I see. Yeah, I'd be interested in fine tuning as well on my own extractions. The tiny model even says it's meant to be fine tuned. I guess that's what they are commercializing :/

Gliner is cc-by-nc-4. Seems like it's not licensed for commercial use?

Where are you getting that it's a closed model? I'm seeing mit license on huggingface.

that's actually the particular one I was looking for and couldn't find. Also had googled for the other ones but maybe it was so recent that it hadn't been indexed. Thanks!

Are these small Gemma 2 distilled models available anywhere? I'm not finding them on huggingface.co, etc. but maybe I don't know the exact model names they are published.

Are the weights released yet?

v0 by Vercel only outputs one UI library, shadcn/ui. Useful if you are wanting to use that library for example starting a project from scratch. Potentially vercel will add more.

https://v0.dev/

Love the upgrade music. Wish more libraries would add upgrade music with major releases