Strawberry is Coming 🍓 New Frontier in Drug Discovery 💊 Drama in AI 📺
What you actually need to know about AI, weekly.
Welcome to this week in AI.
This week, OpenAI’s Strawberry model is in the headlines again. Apple’s new iPhone 16 has some impressive AI features. There is drama in the AI community and much more.
Let’s get into it!
👋 If you’re new here, welcome!
Subscribe to get your AI insights every Thursday (usually).
OpenAI's "Thinking" Strawberry Model to Launch in the Next Two Weeks
OpenAI’s Strawberry model is finally going to be released in the next two weeks.
As we’ve covered before, Strawberry is set apart by its unique ability to "think" before responding to queries, unlike other conversational AI models that provide immediate answers.
This new model is expected to be part of the ChatGPT service but will also be a standalone offering.
Why it Matters
Strawberry's ability to reason will enhance the quality and depth of its responses.
It addresses a key flaw in large language models (LLMs), they are too eager to give a response, which leads to flawed outputs.
An example is if an LLM generates code and you feed that code back to it, it will always try to refactor it rather than *thinking* through if it is the best answer.
The upcoming release of Strawberry is timely, OpenAI is seeking an additional $20 billion in funding.
While the initial version of Strawberry will be limited to text-only interactions, its potential impact on the AI landscape is significant.
We can anticipate even more sophisticated and nuanced AI models in the future.
📰 Article by The Information (paywall)
Apple Intelligence Redefines the iPhone Experience
Apple has just unveiled its suite of AI-powered features for the iPhone 16, aptly named Apple Intelligence.
These new capabilities aim to simplify tasks, provide valuable insights, and elevate the user experience.
Why It Matters
Apple has strategically branded its AI as "Apple Intelligence" to emphasise its practical applications rather than the underlying technology.
These new features promise to streamline daily tasks and interactions, setting a new standard for user expectations in how we use smartphones.
The new AI features:
Visual Intelligence: Use your camera to identify dog breeds, get instant restaurant reviews, and learn more about your surroundings.
Automated Summarisation: Stay on top of your inbox and notifications with concise summaries and prioritised alerts.
Advanced Photo/Video Search: Find specific photos or moments within videos effortlessly using natural language descriptions.
Writing Tools: Improve your writing with AI-powered suggestions and add flair with AI-generated emojis and images.
Enhanced Siri: Experience a more intuitive Siri that understands context and seamlessly executes actions based on your requests.
Third-Party Integration: Access ChatGPT directly from your Camera Control button.
Stanford Study: AI Beats Humans in Generating Novel Ideas
AI is playing an increasingly prominent role in research.
A recent Stanford study has shown that LLMs can generate research ideas that are perceived as more novel and exciting than those from human experts.
The study involved over 100 NLP researchers who participated in both idea generation and blind reviews.
The results were clear: AI-generated ideas were rated significantly higher in novelty (p < 0.05).
Why it Matters
The Stanford study challenges the common misconception that LLMs are incapable of generating novel ideas.
The prevailing belief that LLMs merely regurgitate information from their training data is contradicted by the study's findings.
Despite the observed limitation in idea diversity, with only 5% of AI-generated ideas being unique, the study demonstrates that LLMs can indeed produce novel concepts, even surpassing human experts in terms of perceived novelty.
Fake Models and Drama in the AI Community
OthersideAI and its CEO, Matt Shumer, recently made headlines with claims of achieving a significant breakthrough in AI model training.
The announcement garnered a lot of excitement within the AI community, as it suggested the possibility of tuning models at the *abstract* level, essentially baking prompting techniques to help the model reason before answering.
However, the initial enthusiasm quickly turned to skepticism as researchers failed to replicate the claimed results.
The scrutiny intensified when a private API provided by Shumer, which seemingly demonstrated the model's impressive capabilities, was exposed as a Claude wrapper cleverly disguised to mimic the open-source model.
In other words, they pretended a top-tier model was their own.
Why it Matters
The degree of investment in AI companies right now AI is astounding, Shumer and his associates hoodwinked many in the industry to presumably get funding.
Deceptive practices, such as this, can divert valuable resources and attention away from genuine research efforts.
Fortunately, AI isn’t like other hyped tech (crypto for example), it’s very difficult to fake technical breakthroughs for a prolonged period.
📝 Great X thread on the fiasco
Australian Gov Aims to Regulate AI
The Australian government has proposed new legal requirements for AI in high-risk settings and introduced a Voluntary AI Safety Standard for businesses.
This move aims to address growing concerns about the potential risks of AI while fostering its responsible development and deployment.
The proposed risk-based approach aligns Australia with international standards.
However, there are calls for swifter action, including the introduction of legislation by 2025 and better enforcement of existing laws.
Why it Matters
Governments around the world are grappling with how to mitigate risks without stifling innovation in AI.
By establishing clear guidelines and regulations, the Australian government is creating a framework in which companies can operate, this encourages innovation and indicates a positive stance on AI.
According to the report, the potential economic benefits are substantial, with AI projected to boost the Australian economy by $200 billion annually.
New Frontier in Drug Discovery
Chai Discovery, a startup backed by OpenAI, has introduced Chai-1, a new AI model that predicts the structure of various biomolecules, including proteins, small molecules, DNA, and RNA.
This release has the potential to democratise and accelerate drug development by enabling a deeper understanding of molecular interactions.
Chai-1 stands out due to its impressive performance, surpassing Google DeepMind's AlphaFold3 in specific tasks such as predicting protein-ligand interactions.
Notably, Chai Discovery has made Chai-1's code and tools available for free for non-commercial use.
Why it Matters
The introduction of Chai-1 is a step forward in AI-driven drug discovery and design.
By accurately predicting molecular structures and interactions, this technology could enable the development of new and more effective treatments for a wide range of diseases.
Chai-1 is another key piece of technology which is helping researchers move us closer to a future where cures for major diseases are within reach.
Missed the last one?
That’s a Wrap!
If you want to chat about what I wrote, you can reach me through LinkedIn.
If you liked it, I’d appreciate a share!