leroman

GPT-4 is already hugely useful. If they are able to lower the cost of GPT-4 further, say 5x or 10x that in itself would be useful and huge.

Well we finally have motivation to move upwards in the Kardashev Scale

From my testing the two top models both can do stuff only GPT-4 was able to do (also Gemini pro 1.0 couldn't)..

The pricing for the smallest model is most enticing, but it's not available to me on my account for testing..

Some tasks we are using LLMs for are performing very close to GPT-4 levels using 7B models, so really depends on what value you are looking to get.

I understand this part but it seemed that the 16->8->4 etc is similar to compression of the "net" and seemed to lower quality below 8.

- we have llama.cpp (could be enough or at least as mentioned in the paper a co-processor to accelerate the calc can be added, less need for large RAM / high end hardware)

- as most work is inference, might not need for as many GPUs

- consumer cards (24G) could possibly run the big models

Can someone versed in the ways of math explain how this is different from previous quantization methods?

And specifically, seeing how going from 16fp to 8bit mostly gives same perplexity while anything further seems to lose quality / dumb down the model, how is this even less precise method is able to achieve this?

DigitalOcean has been great, have managed k8s and generally good bang for your buck

Oh yes, I know of one sure way, works 100% of the time:

A man buys an apple, walks around, and sells it. Then, he buys two apples, walks around again, and sells them. Next, he doubles his effort by buying four apples, walking around, and selling them as well. Suddenly, he receives a phone call informing him that his wealthy cousin has passed away, leaving him a substantial inheritance. Now, the man is wealthy.

This is really cool! one use case I can think of is car to car communication to allow them to share some dangers up ahead for example

If we constrain our selves to a pool of words of say Wikipedia entries, minutes names and maybe some other stuff, and use a "super node" like "addition" to kind of act as a math operation.. maybe this makes more sense?

I thought they were using some kind of vector space searches like embedding.. no idea if that's the case

This is where I see K8s fit in, future proofing and un-vendoe-locking (mostly). My own experience running on DO and moving to AWS (free credit ) while using DO S3 clone, managed pg and K8s, the move was rather straight forward..

Yes, maybe the better wording would be "public cloud", still probably everyone reads this as "cloud" vs "private hosting"

Nice, years ago I wrote a TypeScript (originally CoffeeScript re-written to TypeScript) library to generate schedule times for a cron between two time points, should be easy to write a service like this with it.

https://github.com/romansky/JsCron

I haven't got a chance to use Swift but it looks nice!

You can think of generics a meta-language that restricts and documents some behavior one level above the concrete types that you might need.. Lots of benefits, code re-use, documentation as code, less bugs..

Granted it's not always easy and straight forward but that what senior developers in the team are for..

As an extension to this, strong typing is a gateway drug to generics.

Strong typing really shines when you have generics, saving bugs at one level above basic types..

If you created your company via Stripe Atlas, you can get 2K USD credit with OpenAI

Maybe its forward thinking of the ML folk to create big data to train the models that will replace them

For me after trying all the "cool" stuff like mongo etc, seeing the great support Posgres has for JSON just made it an easy choice.

Use all the relational stuff but still have nice native support for JSON and useful extensions.

What else do I need?

Even now, with the addition of vector stores we get PGVector and so it just reinforces my choice over time.

Several years ago while hustling as a single founder, I hoped to get a desk in a private office (anything but open-space), so I would have a place to come to to focus on work, socialize and feel at work.. So I went to see what WeWork have to offer. There I saw a nice floor with nice looking private offices, finally taken to what they thought might be relevant for me, they showed me a tiny space where half a diner table is hung on the wall and over a diner sofa with the second wall behind the sofa, (now-days these types of rooms are used for private 1 on 1 meetings and zoom calls..) for the price of what I would pay a private office of say 2/3 people (pre-WeWork shared office spaces). Just the audacity of the person showing me this for this price struck me how greedy and disconnected WeWork are, got me so insulted I would never check them out again. This is what you get when you get an MBA type business management constantly looking for ways to up the contract prices like its a B2B SaaS business (not that I'm saying this makes more sense there..)

The post COVID era is more stressful for sure.. (pandemic, inflation, war,?)

As a solo entrepreneur trying to bootstrap a business, I find this far from the truth.

While some things may be “easier”, all the marketing, sales and biz dev channels are saturated, “pay to play” as I see it, if you want to sell something.

Have a product? Great! Good luck getting visibility over all these zombie (live until money runs out) but funded startups.

I am able to implement complex logic without using this new "functions" feature, I am considering using it with the agent implementation, but might not need to as the new GPT4 version improves on it's JSON handling so that is already plenty robust for my needs..

Actually, I'm not using Langchain.

I'm now adding "agent" functionality, specifically to enable the AI to do some "research" on the web, at the moment this will also be done without a framework.

So I'm either missing something or I am doing something simple enough that does not require the framework overhead / added value..

Chat UI/UX is a tool for experts. To drive this point home consider a user prompting "produce a founder agreement document", for which the AI will happily produce -something-. Even thought he user is able to read the document he does not understand the contents in "legal" terms. In contrast if the user would go to an expert lawyer, who would start by asking the user some relevant questions (=domain questions) and put together a prompt tailored to the needs and circumstances of the user in the domain language, with all relevant nuances..

I am working to create this experience by augmenting the AI interaction with step-by-step leading questions and interaction UI, similar to how users would interact with a domain expert.

https://pth.ai Would love feedback! :)

I think this is a naive (or cynical?) take on a business, its also their goal to hire and retain both employees and customers.

If the latter is done based on a lie, then you cant align your “commercial purpose” and people who fulfill it..

I’m not sure im reading well into your criticism about user stories, it sounds like you dislike how they are used in your company.

I find user stories critical because they are the lowest common denominator for product/sales/dev/biz etc, everyone gets them, can agree if this functionality will or will not he in release X and pretty much would set expectation correctly between all parties.

Additionally, moving down the dev pipeline, once you have wireframes / more specific designs, you can go over the user stories to make sure they are all supported.

Pretty sure that Israel, for example, is selling it's Dollar reserves due to the strong dollar, to control it's currency. Israel accumulated these reserves during when the NIS was strong to limit the opposite effect..