Taiwan will announce the "Banking Industry Buying and Selling Cryptocurrency" specification in September, allowing traditional banks to provide cryptocurrency trading platforms.

There are currently 27 exchanges under the guidance of the Financial Regulatory Commission.

"The AI Hot 75" startups who are leading the wave of genAI according to NFX.

Nvidia open-sourced a toolkit to address the hallucination issue (/capabilities) of LLMs called NeMo Guardrails. In a nutshell, how it works is that this method uses a database linking to hardcoded prompts, which have to be manually curated.

A study conducted by Mayo Clinic suggests AI could potentially improve time efficiency and standardization for radiation therapy planning in patients with head and neck cancers.

The validation study, published in Frontiers in Oncology, evaluated the efficiency of an algorithm trained by Mayo Clinic and developed in collaboration with Google Health.

Google’s Cloud unit gains key AI chip team to compete with Microsoft

Google has moved the engineering team responsible for making artificial intelligence chips into Google Cloud, a spokesperson confirmed, in a step that could make the cloud unit more competitive with its bigger rivals, Microsoft and Amazon Web Services, in selling AI-powered software to businesses.

The move is the latest sign of how Google is scrambling to respond to Microsoft and OpenAI, whose successful launch of ChatGPT—a chatbot now incorporated into Microsoft’s Bing search engine—threatens to erode Google’s long-standing dominance of search.

Additional evidence highlights how informative could be data collected from eyetracking only, as previously discussed by XRSI - XR Safety Initiative in the context of biometrically-inferred personal data:

"Researchers have identified spatial and temporal abnormalities in spontaneous fixational saccades as a potential biomarker for cognitive and positive symptoms in schizophrenia.

Researchers combined patient data of fixational eye movement recorded over 60 seconds with machine learning technology, which was able to determine schizophrenia with 85% accuracy.

This discovery suggests that fixational saccades could serve as an easily obtainable complementary diagnostic tool for the condition."

Cohere launched Cohere Rerank in beta, a revolutionary new endpoint for document ranking!

CEO IBM Arvind Krishna said the company expects to pause hiring for roles it thinks could be replaced with AI in the coming years.

These non-customer-facing roles amount to roughly 26,000 workers, Krishna said. “I could easily see 30% of that getting replaced by AI and automation over a five-year period.”

That would mean roughly 7,800 jobs lost. Part of any reduction would include not replacing roles vacated by attrition, an IBM spokesperson said.

Walmart is using AI to negotiate deals

Walmart is using a chatbot named “Pactum AI” to automatically negotiate with suppliers.

It saves Walmart an average of 3% on expenses.

Notably, three out of every four suppliers prefer to negotiate with AI rather than humans.

LLaMA spearheaded the age of community-driven midsize LMs

It’s game-changing for many use cases that don’t require GPT-4’s full firepower.

Great blog post that covers Alpaca, Vicuna, Koala, WizardLM, OpenAssistant, and more!

Cognito Therapeutics raises $73M series B to advance neurotechnology platform for Alzheimer’s Disease

New investors Alzheimer’s Drug Discovery Foundation (ADDF), Starbloom Capital, IAG Capital and WS Investment Company (Wilson Sonsini’s venture arm) joined the round.

The Series B funding brings the total amount raised since inception to $93 million.

Excellent TED talk from Sal Khan:

- many inspiring examples of GPTs finetuned into socratic tutors, assisting without giving away answers.

- none of it "out of the box", requires prompt engineering, finetuning, data collection, iteration.

- sense of barely scratching the surface.

Inflection AI, the startup from ex-DeepMind cofounder Mustafa Suleyman, has launched its first chatbot, Pi > https://heypi.com/talk

It’s designed to be better at natural, flowing conversation than lists, plans, or code.

NVIDIA’s groundbreaking graphics research

NVIDIA released that they will present around 20 research papers at SIGGRAPH 2023.

In one demo, they create realistic hair for 3D characters in real time.
It seems advanced technology is putting an end to bad hair days.

Research papers range from AI models that turn text into images to inverse rendering tools that transform images into 3D.

NVIDIA’s research has been on fire lately.

In Israel, AI surveillance is being used in ways comparable to what China did in Xinjiang.

Facial recognition reinforces boundaries in the West Bank w/ traffic light color codes. Soldiers train algos at checkpoints, collect Palestinian biometrics w/ apps.

Most people haven’t had a chance to use H100 yet. Theoretical FLOPs mean nothing - true speedup on Transformers is all that matters.

How does H100 fare against A100 in real battles?

Results on GPT from a third party MosaicML: the larger model you train, the more time you save.

Easiest way to get 32k context GPT-4 model access is to submit a good open source Eval that’s accepted

Evals that creatively utilize long context will be prioritized for review.

Submit your Evals to this repository
https://github.com/openai/evals

Microsoft is working with AMD both on Microsoft's own AI GPU, code-named Athena, and on bolstering AMD's own efforts to become a second supplier of GPUs apart from Nvidia.

To be clear, Microsoft is still very much relying on Nvidia and is in fact also working on how it can get more Nvidia GPUs because Microsoft doesn't have enough for all of its needs right now.

But the idea is for Microsoft to develop multiple options to get more GPUs and perhaps at some point to have some less expensive options to Nvidia. Of course for AMD, the idea is to leverage the AI demand boom to build a big GPU business.

Deepmind CEO says we'll have AGI in a few years

Deepmind CEO "progress in the last few years has been pretty incredible...I don’t see any reason why that progress is going to slow down. I think it may even accelerate. So I think we could be just a few years, maybe within a decade away.”

What programming language will run AGI?

Well, it might be Mojo. Modular just released a Python-like programming language that runs at the speed of C!

• *Truly* parallelizable
• Fully compatible w/ all Python packages
• Speed boosts from types (but not required)

Samsung will overtake TSMC within 5-years, the president of Samsung’s chip division said. “To be honest, our foundry technology is 1 or 2 years behind TSMC’s.” But once TSMC starts making GAA transistors at 2nm, “Samsung will lead the way,” he said.

Note: Samsung entered the chip foundry business 18-years ago and still lags behind TSMC.