emadm

I’m building a system for free universal ai access for the important things in life - education, health, government etc

Stuff that should be open source, open data

Made state of the art datasets, health models, research systems & agents so far @ www.ii.inc but the plan is ai first open source full stack systems for every regulated sector

Have a distributed ledger announcing soon to tie it all together and create a flywheel so more folk can get access to ai

Schelling AI | Interns | London, UK

Open source, decentralised generative AI launching later this month from original team at Stability AI.

Hiring an intern class working on a series of pieces on how generative AI will impact sectors, society and more, plus how the technology will evolve.

This would particularly suit those passionate about AI & Web3, working direct with CEO and researchers based on location in London (High Street Kensington).

Apply at careers@schelling.ai with the most original take on how generative AI will impact society or a sector you can think of plus CV.

"The increasing amount of time Altman spent at OpenAI riled longtime partners at Y Combinator, who began losing faith in him as a leader. The firm’s leaders asked him to resign, and he left as president in March 2019. Graham said it was his wife’s doing. “If anyone ‘fired’ Sam, it was Jessica, not me,” he said. “But it would be wrong to use the word ‘fired’ because he agreed immediately.” Jessica Livingston said her husband was correct. To smooth his exit, Altman proposed he move from president to chairman. He pre-emptively published a blog post on the firm’s website announcing the change. But the firm’s partnership had never agreed, and the announcement was later scrubbed from the post. For years, even some of Altman’s closest associates—including Peter Thiel, Altman’s first backer for Hydrazine—didn’t know the circumstances behind Altman’s departure. "

From the Wall Street journal

https://archive.is/WiqtZ#selection-1177.0-1207.168

Made that 4m last a long time eh ^_^

I figured it would be interesting to see scale and for the image models

This with audio to audio, style transfer & all the elements to come likely put into a comfyUI style format is just the beginning as a inspiration and augmentation tech.

Team aren't that interested in Taylor Swift by Drake or whatever but steadily building up tools to customise to musicians own libraries (the open version of this, it tunes super well), plus then have control and composition over it all.

Plus making sure the licensing is all good versus other audio tools where they are asking for permission later (which is a very bad idea imo)

I started as a hedge fund manager at 23 doing global macro at Pictet & quit when my son was diagnosed with autism to build an NLP team to analyse clinical trial & other data and undertook pathway analysis of neurotransmitters focusing on GABA/Glutamate balance to repurpose medicine to help him filter, after which he went to mainstream school.

One of the things I am focused on now is bringing that research with the new medical LMs so for autism, cancer, MS, Alzheimer’s & more everyone has comprehensive, authoritative and up to date knowledge at their fingertips on these and also a guide so they never feel alone and helps them on their journey if they or a loved one gets this diagnosis.

It will be open and transparent, took a while to build to here but looking forward to next steps.

There is more to life than power and money.

I was lead architect of CAIAC to organise Covid knowledge launched at Stanford before this

https://hai.stanford.edu/watch-caiac

The big companies here and more promised me tech to build the system and didn't deliver which is why I did Stability to make sure there were open alternatives.

Still achieving the output we did based on completely diverse global team form mostly untraditional backgrounds outside SF & the resources we had is a positive not negative story.

The first 20 employees I hired were graduates from the job centre in Ilford and we built this above a chicken shop.

Talk about that here https://x.com/PeterDiamandis/status/1773863016775213545?s=20

Company has money, revenue is better than peers with a path to profitability and one of the fastest ramps ever.

Tesla, Apple, NVIDIA all had these moments as well as innumerable other good companies.

Was tough to navigate through, very proud of team.

To be clear: we never raised less than $25m of additional funding.

Here are some examples of the work that I actually did in public that they chose to run as I said I was a spy

https://www.wsj.com/articles/emad-mostaque-the-spreading-men...

Here is me predicting ISIS as they would later be known would likely topple Western Iraq the following summer in December of 2013, top of business insiders chart of the year (which I think is very interesting)

https://www.businessinsider.com/most-important-charts-2013-1...

Bloomberg article is better, but note they also ran that I said was spy for the British government when I worked in counter extremism work including writing in this topic for Reuters, Wall Street Journal and others.

These are details we gave them and they chose not to run.

In addition they ran reporting that we raised less than $25 million of additional funding when we raised well over that when we said that was false based on sources again.

We had to take it legal and right up to lead editor.

I kinda gave up on these reporters when they insisted on posting obvious lies, like in this case I don’t speak to Jensen since October 2022 and numbers and other reports were clearly fabricated.

Like what do you do?

I think you just ignore and ship good things, some great models coming across modalities and Stability AI is one of the fastest growing revenue companies ever despite the BS against us

https://x.com/emostaque/status/1773766734522040508?s=46

https://pytorch.org/blog/high-performance-llama-2/

Yeah they train well and very stably even int8, maxtext now has LLaMA and mistral support too, pytorch xla gets 50% MFU with spmd and you have some nice stacks like levanter

Haven't been too impressed with inference versus tensor rt llm for example though

Yes, this is possible, we have ComfyUI workflows for this.

Yeah will all be downloadable weights. 800m, 2b and 8b currently planned.

where is Ilya

Should look at MLX optimisations too, Stable LM 1.6b which is about the same size and quantises to 4bit really well runs 100 tok/s on a M2 Mac mini.

https://x.com/awnihannun/status/1750986911827832992?s=20

https://stability.ai/news/stablelm-zephyr-3b-stability-llm works absolutely fine on the M2 processor, like 40 tok/s https://x.com/EMostaque/status/1732912442282312099?s=20

Stable LM 2 1.6b runs even faster but not as good at RAG, great multilingual though, we are seeing it matching 70b models on other languages (new version soon) https://x.com/EMostaque/status/1763269238347673796?s=20

Can fit a lot in a gigabyte file it seems.

Yeah we were surprised by that, video models are great 3d prices and image models are great video model priors

Our team at Stability AI build ComfyUI so yeah is supported

Heh https://stability.ai/news/stable-zero123-3d-generation

Better coming

We are starting to see that, see phi2 for example

There are approaches to get the right type of augmented and generated data to feed these models right, check out our QDAIF paper we worked on for example

https://arxiv.org/pdf/2310.13032.pdf

Yeah chain it in comfy to a turbo model for detail

We did try stableLM 3b4 with books3 and it got worse in general and benchmarks

Just did some pes2o ablations too which were eh

Law for music is different to other media types

Different deal with the underlying data holders with revenue share etc

a16z are investors in openai

Sd turbo runs nicely on a m2 MacBook Air (as does stable lm 2!)

Much faster models will come

If you try diffusionGPT with regional prompting added and a GAN corrector you can get a good idea of what is possible https://diffusiongpt.github.io