OpenAI’s Q* Rears Its Head Again 🍓 AI Predicts Alzheimer's with 80% Accuracy 🧠 Meta's GPT-4 Killer 🚨
Your weekly dose of AI insights
Welcome to this week in AI.
This week - OpenAI’s Q* rears its head again, this time under the codename Strawberry. More scientific progress in the form of a Ahlzeimer’s disease diagnosis and Meta threatens to unleash an open-source rival to GPT-4.
Let’s get into it!
👋 If you’re new here, welcome!
Subscribe to get your AI insights every Thursday.
OpenAI's Project Strawberry: A Leap in AI Reasoning?
OpenAI is reportedly developing a new AI model, under the name Project Strawberry, with the potential to significantly enhance reasoning capabilities in AI systems like ChatGPT.
This model, previously known as Q*, aims to enable AI to perform deep research, autonomously navigate the internet, and complete complex, multi-step tasks over extended periods.
The project has garnered significant attention due to its potential to accelerate progress towards artificial general intelligence (AGI), which aligns with OpenAI's core mission to create AGI that benefits all of humanity.
However, the project has also sparked concerns as Q* reportedly played a part in the brief ousting of OpenAI's CEO Sam Altman in 2023.
Prominent figures like Elon Musk are raising the alarm, Elon tweeted, "We were told AI doom would be a paperclip maximizer, but it’s actually going to be Strawberry Fields Forever".
Why it Matters
Project Strawberry's development marks a significant step in AI research. Its success could lead to AI systems that are more independent and capable of solving complex problems and lead to a new era of autonomous AI agents that act on our behalf.
This could have far-reaching implications for industries ranging from healthcare and education to engineering and research.
For comparison, the current leader in reasoning and math capabilities, Claude Sonnet 3.5, achieved a 71.1% score on the MATH benchmark test (a dataset of championship math problems). The Strawberry model, meanwhile, reportedly scored over 90% during internal testing.
This significant difference in performance suggests that Strawberry has the potential to surpass current capabilities in these areas.
AI Tool Predicts Alzheimer's Progression with 80% Accuracy
Researchers from Cambridge University have developed a new AI tool that boasts over 80% accuracy in predicting whether individuals with mild cognitive impairment (MCI) will progress to Alzheimer's disease (AD).
The breakthrough leverages data from cognitive assessments and MRI scans, eliminating the need for invasive procedures like PET scans or spinal taps.
The AI not only identifies those likely to develop AD but also categorises patients into groups based on their risk of progression, aiding in personalised treatment plans.
Notably, the tool has been validated across multiple memory clinics internationally, highlighting its potential for global application.
Why it Matters
Early detection of Alzheimer's is crucial for effective intervention and treatment.
This AI tool could help dementia care by enabling early diagnosis and personalised treatment strategies.
Its non-invasive nature makes it more accessible to patients, potentially leading to widespread adoption in clinical practice.
With the number of dementia cases projected to triple in the next 50 years, this AI-powered tool offers a beacon of hope for timely intervention and improved outcomes for individuals facing cognitive decline.
Meta’s Most Powerful Model Yet
Meta is gearing up to release its largest and most powerful Llama 3 model yet, with a huge 400 billion parameters.
This new model is expected to rival the performance of OpenAI's ChatGPT-4 while being more efficient in terms of computational resources and cost.
The model's enhanced capabilities, including multimodality (understanding and generating various types of data) and the ability to converse in multiple languages, make it a real contender.
Importantly, Meta may release this model under an open license, making its advanced language capabilities freely available to researchers and developers.
The model is forecast to be launched in late July or August 2024.
Why it Matters
The open-source nature of Llama 3 400B could democratise access to cutting-edge AI models, fostering innovation and accelerating the development of novel applications.
Meta's strategy here seems pretty obvious: by releasing Llama 3 400B under an open license, they're making it difficult for any one company to build a moat or successful business around large language model supremacy.
This could lead to a more level playing field in the AI market, fostering more competition and potentially faster innovation.
OpenAI Exits China, Opening Doors for Local AI
OpenAI's recent decision to block access to its tools and services in China has sent ripples through the country's AI community.
While the US firm has not provided specific reasons for the move, it is likely linked to rising US-China tensions and restrictions on exporting advanced semiconductors crucial for AI development.
This decision has left Chinese developers scrambling for alternatives, creating a golden opportunity for domestic AI companies.
SenseTime, Baidu, Tencent Cloud, and Zhipu AI have swiftly responded by offering attractive incentives, including free tokens and migration assistance, to entice OpenAI's former users.
The competitive rush signals growing potential and vibrancy of China's AI market, which boasts an estimated 130 large language models, accounting for 40% of the global total.
Why it Matters
OpenAI's exit, while initially disruptive, may ultimately accelerate the development of Chinese AI and the global industry through competition.
With increased market share and a renewed focus on innovation, Chinese AI companies could bridge the performance gap with their US counterparts.
This shift could lead to a more self-reliant and robust Chinese AI industry, fostering greater competition and potentially driving advancements in AI technology on a global scale.
Rep. Jennifer Wexton Finds Her Voice Again Through AI
Rep. Jennifer Wexton (D-Va.), who has lost her voice to progressive supranuclear palsy, has debuted an AI-generated version of her voice, created by ElevenLabs.
This technology has allowed her to continue participating in her legislative duties and public engagements.
The AI voice, which sounds remarkably like her own, has restored her ability to communicate effectively, drawing an emotional response from Wexton and her colleagues.
This breakthrough was made possible through a collaborative effort by Wexton's team, which provided audio clips, enabling ElevenLabs to produce a natural-sounding voice model.
Why it Matters
Rep. Wexton's use of AI to regain her voice highlights the transformative potential of technology in improving the lives of individuals with disabilities.
This advancement not only enables Wexton to perform her duties effectively but also underscores the broader implications for AI in assisting those with speech impairments.
📰 Article by the Washington Post
Exposed: How AI Companies Are Secretly Exploiting YouTube Videos
A recent investigation by Proof News has revealed that several leading AI companies, including Apple, Nvidia, Anthropic, and Salesforce, have used subtitles from over 170,000 YouTube videos to train their AI models without obtaining permission from the creators.
The dataset, known as YouTube Subtitles, contained content from a vast array of channels, ranging from educational institutions like MIT and Harvard to popular creators like MrBeast and PewDiePie.
Why it Matters
This unauthorised use of copyrighted material raises significant ethical and legal concerns.
Many creators were unaware that their content was being used and have expressed concerns about the potential misuse of their work and the lack of compensation.
This revelation highlights the need for greater transparency and regulation in the AI industry, as it grapples with the challenges of balancing technological advancement with ethical considerations.
It also underscores the importance of respecting intellectual property rights in the digital age, particularly as AI technologies continue to evolve and become more sophisticated.
Bonus Videos - New SORA Showcases
That’s a Wrap!
If you want to chat about what I wrote, you can reach me through LinkedIn.
If you liked it, give it a share!