applgo443

What's the simple explanation for why these VLM OCRs hallucinate but previous version of OCRs don't?

Why are traditional OCRs better in terms of hallucination and confidence scores?

Can we use logprobs of LLM as confidence scores?

When i deploy a webapp on azure, it expects me to put env variables in a file or their own tool (key value fields) where you can add env variables one by one.

Is there a way to use envelope in places like those?

If it's 5 common crawls, isn't data across multiple common crawls mostly similar?

Is ETL/ELT same as writing SQL scripts and periodically executing them? I assumed there's more to it.

May be rewriting user's description might help you match code better?

Similar to the prompt engineering for previous era GPT completion models.

How do you approach the problem of what files to look into to fix a bug? Just embeddings doesn't seem to cut it.

How is your experience with Modal?

And I'm curious to know more about your costs of deployment and running on Modal.

Did you consider first asking LLM to explain what a code snippet does and use that instead?

It'd significantly increase the costs though.

I saw your comment, got curious, and looked at a lot of your old comments. Lots of interesting insights - Thanks for sharing them.

If you don't mind me asking, what do you do? I'm a researcher at FAANG working on language models and starting a new company in the space. Would love to connect. Feel free to email me - idyllic.bilges0p@icloud.com

Used OpenAI APIs before but not function calling. What's the use of function calling in this example?

How does the confidence scores work?

I’m a researcher in the space exploring few ideas with the intention of starting up. Would love to reach out to you and talk to you. Is there a way I can contact you?

My email is beady.chap-0f@icloud.com

I think we can detect atleast a few things like PII leaks etc. Don't you think those things alone are valuable?

They mention the contextual stack is is relatively underdeveloped. Any idea on what can be improved there?

What do you mean by firewall layer? What tools do you use here?

Wow, if you're competing with ChatGPT, that gives all other startups out there hope that competitors aren't something to be super scared about :)

Location: SF Bay Area

Remote: Preferred

Willing to relocate: Depends on opportunity

Technologies: Python, Pytorch, Tensorflow, Jax, machine learning, deep learning, language modeling,

Email: beady.chap-0f@icloud.com

I'm a Research Engineer at Apple, working on Language Models. Have expertise in both building LMs + using these LM APIs to build apps. I'm adept at the fundamentals of machine learning, deep learning and worked across NLP (language models), NLP + CV (Image captioning), and reinforcement learning (PPO for image captioning, robot navigation, etc). I am also a proficient C++ systems developer and worked on several low level systems problems such as sorting, caching, etc before I entered ML space.

What I'm looking for: Though I'm primarily a research engineer, I'm interested in joining a place that lets me wear multiple hats, and gives me enough responsibility and opportunity to scale my personal learnings and skills along with that of the company. My mid term goal (~ 5 years) is to start my own company. Hence, for my next opportunity, I'd like to join a small startup where I can also learn the non-technical parts of building a company.

Why are there so many weather APIs? What's going on here?

I considered doing this - take screenshots of your screen constantly, OCR them and index them. It's fairly simple. However, there are some problems

- OCR constantly running in the background is power consuming - What granularity do you take your screenshots? Imagine each screenshot is 500 Kb and you take one each second. This'd result in 40 gigs of data per day. How are we gonna store it? How many days data do you want to keep?

Man, this is exactly what I did at a previous job. We had a massive Query engine as a part of our BI product. I've recoded the entire sorting algorithm to use variadic templates to get a sorter for any set of column types in one go, rather than dereferencing a pointer for each column. We've observed massive improvements in speed (more like 50%, rather than 100x mentioned here. However, that codebase has been constantly optimized over the last 25 years).

I work on building language models at FAANG for my day-day job and I'm very curious about this - how does finetuning on T5 remove hallucination? Can you elaborate on this?

This is an amazing product, btw. Let me know if you're looking for people to hire :)

How do I trust that these editors are good? How do you get these editors?

And what do the editors specialize it? Can I write cooking blogs? Literature blogs? Computer science paper reviews?

Can you show me a post before editing vs after editing? Only then I'll understand the value provided.

What does this mean?

"If founding a technology company is a dream of yours—even if you don’t yet have a fully formed idea and haven’t yet quit your day job—we want to hear from you!"

Why can't we use this quadratic convergence in deep learning?

What do you want?

Interesting.

How do you figure out which are blogs and which arent'?

This is amazing!

A meta rant - There's a lot of cancel culture at play around working hard.

I don't understand why people are trying to guilt/shame others into working less. And yet, at the same time, watch documentaries about successful people (Kanye, Michael Jordan) talking about how hard they used to work and celebrate it.

In order for you to be on the top of your field, you need to put in hours and hours of work. You'd have to sacrifice the so called 'life' part of the equation. If you don't want to do that - fine! We just don't share the same values and priorities. Just like how I shouldn't shame you for not working hard, you shouldn't shame me for working hard.

If I start a company, it's totally okay for me to only get employees who share this view. I'd boast about how hard I work and how much we accomplished. I'd let my employees do that. And I'd compensate them accordingly. If you think that's toxic, so be it. I feel the real shame is not realizing your full potential and falling short of your dreams.

This is amazing! Do you also think the quality of the item delivered would be the same? Despite whatever website we go to? I know for a fact that 'same' electronic items sold at walmart are cheaper and are of slightly lower quality than those sold at home depot, etc.