lee101

I built a web crawling site ring.nz on polymer/web components. Looking at the marketing of the framework it was an easy way to get material design components but in reality the overuse of shadow Dom and shadow CSS meant it was really difficult to use the components or theme them. I'm pretty sure the whole polymer project is abandoned now... The site became too hard/bothersome for me to maintain under its tech debt but is still on my GitHub.

I also regret building some things in C# or Java a long time ago.

The biggest regret money wise is I built a crypto forecasting company that became somewhat profitable BitBank.nz but I built it without enough foresight for cloud costs, was running a forecaster constantly on compute engine as well as filling up a postgres database without offlining data, turned out harder to scale down a database than scale up the memory. Basically things like that and I never had time for it after having kids meant it long term costed more than it made even with paying customers so I shut it down.

Now I run https://text-generator.io still machine learning related but I run it very lean from two GPUs I own at home which is a bit easier

Also might be just SEO squatting... May throw up a bunch of AdSense ads on their side, the fact that there is a duplicate site is a problem for you, do what the comments say with adversarial stuff and complain to as many parties as possible including DMCA, registrars and the Microsoft/Google deceptive Site lists, etc. Don't let your domain expire or you'll get it bought of you and content mirrored there too which is surprisingly common

False dichotomy.... After being at many successful and unsuccessful companies there's definitely different vibe, on one side you as an individual are allowed creative freedom to design+deliver a large project... on the other theres a lot of fear blame, time sunk into these "core" projects eg going mobile or video First or changing Clouds or cost cutting.. they often turn out to be illconsidered wastes of time, the pretence for war is a poor excuse for this lizard brain/ groupthink/ just doing what said manager likes to do

Checkout my business for AI text generation https://text-generator.io !

Apologies for owning said domain, I worked at a domain name registry... Attesting to the other comments that it's a metaverse/nft/digital property style pyramid scheme like most things and they do tend to be owned by "evil capitalists who do hostile takeovers".

Even the web3 name tokens I'd assume there's someone who minted a whole bunch of tokens for themselves before launching.

You can buy names that are no dollar markup on CloudFlare which still gives money to the registries but at least that skips the bottom layers of the pyramid

Speech to text is still really bad, Cloud costs are real bad too, checkout https://text-generation.io or OpenAI whisper which will fix that fairly soon I think. Over 8x cheaper than Google.

There's lots of other cloud cost stories like CloudFlare R2 egress saving people millions on large file storage.

I'm surprised decentralization hasn't hit cloud providers as hard yet either.

With that hotspot example my Telco in NZ (spark) blocks me from using a Wi-Fi hotspot on one of my android mobiles, which is ridiculous that they can turn off an Android feature on my phone for profit.

This is already happening, check out https://text-generator.io around 10x cheaper than OpenAI for text-generation also has speech-to-text 8x cheaper than Google cloud.

New era of cost-cutting, running systems yourself and distributed computing will bring far cheaper prices soon. If we do well energy and computation/intelligence will be trending toward 0 generally speaking

for now I'd checkout T0 or bloom, lots of models can be instructed already.

I created https://text-generator.io and have a whole bunch of examples on there of instruction following style prompts

I'm running https://text-generator.io from my house, two 3090s right now powering it. It allows the service to undercut OpenAI around 10x on Text/code generation and Google over 8x on speech to text. A Cloudflare tunnel is pointing to it running locally. It makes development very fast too. Its a bit tricky to keep purchasing new hardware to spin up new instances but that's getting easier with practice and autoscaling cloud providers doesn't necessarily work that well either.

I think co's should seriously consider this or at least adding everyones development machines to the prod cluster during when they sleep, which is what we did to render movies when i was at Weta Digital. 1000's of developer machines are pretty valuable if put to good use.

Cheap competitors to gpt-3 are poping up like https://text-generator.io

Lots of the startups can pivot easily to use these cheaper systems after capturing the market with OpenAI powered apps, especially as it's just a one line change https://text-generator.io/blog/over-10x-openai-cost-savings-...

Or train their own algorithms etc so there will be lots of winners not just OpenAI and the tech Giants

Also checkout https://text-generator.io which can do the same things but much cheaper, it can also analyse text in images e.g. if it's an screenshot of a C.v.

Nice, mosaicml is a great library for training with modern enhancements like alibi which really helps.

Also for inference checkout https://text-generator.io doesn't do training but is extremely cost efficient if you can just prompt tune your way to something

Checkout https://text-generator.io which can also be made to translate between human and computer languages with prompt engineering, has a nice 100 requests a month free teir

I made https://text-generator.io basically to bring down the costs of ml/ai, people enjoy it because it's around 10x cheaper than OpenAI and does multilingual text and code too, now I'm doing speech-to-text 8x cheaper than Google so saving people a lot of money right now. Hopefully it's meaningful already but I'd like to bring the cost of any intelligence down to near zero which would I think make a big difference having things all operate well for us.

My mum was a punch card and fortran programmer, my brother's a software engineer too so it's in the family I suppose, i got into it making games and I got an apprenticeship making websites out of highschool, that set me up to go through computer science faster and easier than normal in 2.5 years

Also check out https://text-generator.io which is over 8x cheaper than Google for speech to text, 5.5 hours free every month too

I've been enjoying FastAPI after using flask a lot, there's a bit of learning like making routes async and pushing background tasks for later. But I think it's good

Yea, images/video is interesting, I think combining textual instruct and dalle/other models would be interesting so you could describe literally "restore this old photo and upscale+colorize it".

Was thinking of ways to do that kind of thing, I think unified IO is a good start at some flexibility but still lots of peices missing.

In the world of video training your own video filters based on editing a few frames is quite common E.g. something like nuke copycat or using ebsynth etc.

There's likely some new wave of DSLs or visual workflow builders coming that are either more ml friendly and or specific to building a conversational system/ image editing pipeline etc

Right now the text generation API can be used to do the conversational stuff along with a massive amount of other stuff like classification or adding questions about what is in text or how proper would respond to things for analysis purposes.

There's an embeddings API for text/code/images that can bed used for the similar snippets stuff if you bring your own vector search engine like milvus etc, but eventually I'll support uploading sets of documents and do the finding relevant snippets behind the scenes so you can ask these kinds of questions e.g. so people can do custom site specific FAQ answering bots etc.

There's an example in the playground of shorter autocomplete which is another way of helping agents too that it supports

Thank you! Means a lot :)

I run https://text-generator.io competing with OpenAI. I used to use kubernetes on Google cloud, but now run it from my house behind a CloudFlare tunnel believe it or not Using my own GPUs allows me to undercut them with around 10x cheaper pricing.

The product is built with some open source and some retrained models, it analyses input links or linked images with tesseract/unified IO / OFA / beautiful soup so it can answer questions about images/receipts etc.

The site itself is on app engine standard python with stripe/firebase UI for auth, which is running very cost effectively.

In future in planning to layer on more content understanding eg audio with whisper etc, then milvus/haystack for lookup phases for more factual question answering, then some work on automated retraining and regeneration etc for hard cases, then searching other systems/internal code and logic execution etc, eventually with a few more steps I'm hoping it will feel like I made AGI in a rack in my house lol...

Nice, for images or web links checkout https://text-generator.io there's a blog there about how it crawls images https://text-generator.io/blog

Hi, love the app! I'm founder of https://text-generator.io

i'm working on my own text generation API, it already does a lot of what OpenAI GPT/Codex do in that it can generate multiple languages and code. It also does more in that it understands web links and images.

I'm working on question answering (with milvus/haystack) right now so really interested in how you guys did it.

Great product, checkout https://text-generator.io anyway as it can save you LOTS of money on GPT-3

This is really interesting, i think python itself is a bit too tricky or easy for a machine to get wrong (indentation etc can come out wrong), i was thinking of trying something i call "eval nets" where you create and train on a dataset with eval in it e.g. "i have 10 apples and need to give them to 3 people so each get eval(10//3)" then evaluating that behind the scenes.

I run https://text-generator.io so i would host it behind there. There's also the options of dropping other networks in for specific tasks like sequences or googling for things with google("blarg") etc. Basically we need some new DSLs for things we want to teach.

Nice, also checkout https://text-generator.io which generates code via API

Yep its really painful, i made https://text-generator.io and considering tackling speach to text for this reason that writing myself is still painful, AI tools help and i'm using https://text-generator.io to help me type (autocomplete) which is a lot better, but this predictive nature needs to be combined with speech to text.

It might feel a bit cumbersome typing autocomplete and talking at the same time but i think its possible with low latency, the latency of the speach to text in the mobile phone is also a real problem, i cant see that its typing anything so i'll repeat things or it will put stuff in there like "hey google". or it gets injected with something in from the radio. Its really plain bad and there needs to be a tool for pros made, i think there will be something made this year id bet

Also if your tired of counting input and output tokens, https://text-generator.io has billing per request and a free tier which is a lot more manageable

In working on a text/code generation system, similar to OpenAI gpt/codex it's a bit different in that it researches links via crawling and understands any linked images. Also different in that it has reasonable pricing and a free teir.

https://text-generator.io/blog/text-generator-now-researches...

It also does embeddings for text/images and code. The problem i suppose is general agi where you have unstructured input and unstructured output and you need to do a lot of research in the process of producing the output which one day may include training, using other web services/information retrieval etc.

Looking to expand the system to understand more and more unstructured inputs and be able to output more and more advanced outputs like images, audio, video, datasets etc.

I was using OpenAI a lot and realised it's going to be dang expensive which is why I started https://text-generator.io so more people can use it

I'd be interested to see if we can work together in some way, i'm founder of https://text-generator.io which also crawls links to the web/images to generate better text: https://text-generator.io/blog/text-generator-now-researches...

Some people self host it, so it can solve some subset of the questions, the image understanding part is important, both it can understand a question with image e.g. if a user drops a fridge to search by image, or multiple images (e.g. which of the 10 images is the nicest looking fridge in 1 API request) as well. Also supports getting shared embeddings for images/text/code, which can be important for the information retrieval/question answering example where it needs to first find the relevant context on wikipedia then feed to the reader model to read it out

Also do other custom stuff like retraining etc. Thanks, Lee https://leepenkman.appspot.com/

Awesome, I wonder if the app is recording what we do so it can replicate what we do, or maybe if not it should have a training mode where we tell it what we do then do it so it can learn.

I feel like some of this could one day be built using a shared model that understands HTML and JavaScript code etc with a few example prompts. Or maybe something that understands intent+a browser automation language like Selenium, if not then some custom input output language+training as adept alludes to.

If interested in building something like this also checkout https://text-generator.io which already pulls down links and images to analyse to generate better text so has a lot of the required parts

I called my startup https://text-generator.io despite the name ... it uses image recognition to help generate text and offers an embeddings API, trying to extend it now to generate other forms of content (images, audio etc) depending on the input which is unfortunate

Running https://text-generator.io which is a cheaper alternative to OpenAI text generation, with a few improvements like it also researches links and understands linked images via analysing any image content, also you can generate a set number of sentences or stop based on hitting low probability words (enables auto complete).

Graphics cards cost do cost around 500 a month though right now which makes ml pretty tight margins... Hoping to sell it more to people wanting to self host in future for that reason