OpenAI Releases Strawberry !

Good morning.

***BREAKING*** OpenAI’s new reasoning engine is out!! See more details below. We talked about this engine extensively in the past here and here. But now - so I don’t have to throw away all this writing for a few minutes, back to the original programming !

A few weeks ago, we compared using ChatGPT to having Einstein in your basement.

This week I needed "Einstein's" help in figuring out a simple planning dilemma.

Every few months, three of my friends (yes, I have friends) and myself go on a hike. We walk from point A to point B but never in a circle. So we need a car at the end point and a car at the starting point. For this week’s hike, some of us needed to get home earlier, so I asked ChatGPT to help reason through the planning.

Who puts his car at the starting point and who puts their car at the finish, and oh boy...

This was not Einstein that was helping me, but Einstein with brain damage and an innate desire to please me.

What should’ve been a simple conversation ended up in a 2-hour back and forth where “Einstein” could simply not grasp the concept that when a person drives his car from point B in the morning to starting point A , that when this person WALKS to point B HIS CAR IS NOT LONGER AT point B but at point A.

If I had hair to pull on, I would have pulled it out.

And when i pointed it out to the chatbot, it was like, Oh yes, of course, sorry, here’s the solution. And then he proceeded on making the same mistake. Over and over again. It was quite interesting. Dumb as a rock.

It was a good case study in aid of those people that claim that Large Language Models are a party trick. There is no intelligence there, no reasoning, and certainly not consciousness and sentience. It’s a dumb computer that is nothing without us.

This is why investors in AI are getting nervous. All that money, and where are the breakthroughs? Where is the promised land of milk and honey where we can use AI to do our bidding as we sip Mai Tai’s on the beach.

Is this the biggest scam in the history of scams ?

But here at the Blacklynx Brief we’re believers, and we think that perception is about to change.

Why, Because it seems now that OpenAI is not only working on feeding their models new and more data (which should make them better), but they are also now working on the reasoning capabilities of their models.

Enter Strawberry or Q*—if you have a paid ChatGPT account, it is now available

This one SHOULD be able to solve the little dilemma I had this week (test incoming)

Remember, this is the model that allegedly was able to solve a 256-bit cipher and starting out knowing anything about math. This was the model that caused OpenAI to fall into two camps.

They’ve hopefully put guardrails on it but still - this should be a step closer to AGI or Artificial General Intelligence.

So let’s hope it knows that when you walk away from your car - you’re no longer in your car (eyeroll).

Stay tuned for next week as we’ll subject this to some rigourous testing in the course of next week!

By the way : there is now a referral program for this newsletter. For every 3 people that you get to sign up we will gift you a 15€ Amazon gift card.

Scroll to the bottom of the newsletter for details !

AI News
New iPhones and exciting developments in self-driving.

• Google DeepMind unveiled AlphaProteo, an AI system that designs custom proteins to bind with specific molecular targets, significantly accelerating drug discovery and cancer research. AlphaProteo creates strong-binding proteins more efficiently than current methods, potentially saving billions in R&D costs and speeding up the development of targeted treatments for diseases like cancer and COVID-19.

• Replit launched Replit Agent, an AI-powered tool that enables users to create software applications from simple text descriptions, making coding more accessible. The tool automatically generates plans, chooses technologies, and builds web-based apps, allowing even beginners to create fully functional applications with ease.

• Researchers developed MarioVGG, an AI system that generates an infinite version of Super Mario Bros gameplay without a traditional game engine. Trained in just 48 hours, MarioVGG can simulate game physics, create new levels, and provide continuous gameplay, highlighting a future where AI can generate custom interactive games based on simple text commands.

• Weave announced Isaac, a versatile personal robot designed to handle household tasks like tidying, folding laundry, and organizing spaces, with a planned release in fall 2025. Priced at $59,000 or $1,385 per month, Isaac is part of the growing competition among companies like Tesla, 1X, and AGIBOT to develop affordable, intelligent home assistants.

• A new Tesla patent reveals an advanced wireless charging system that could allow autonomous vehicles, such as Robotaxis, to charge without human intervention. This technology, potentially being unveiled next month, could enable self-driving cars to operate almost continuously, enhancing Tesla’s autonomous fleet capabilities.

• Researchers developed AI agents to play and evaluate procedurally-generated video game levels for engagement, helping speed up game design. The AI accurately rated engaging levels higher than unengaging ones, indicating that AI could soon help in automatically testing and creating more enjoyable games for human players.

• Apple announced the new iPhone 16, featuring the A18 chip and a full suite of Apple Intelligence (AI) features designed to enhance user interaction. With AI tools for writing, photo and video searches, and improved Siri capabilities, Apple is integrating generative AI directly into its devices, positioning itself ahead in the AI-driven smartphone era.

• OpenAI and Thrive Capital backed Chai Discovery, an AI biology startup that raised $30 million to develop AI models for drug discovery. Chai’s AI model, Chai-1, claims to outperform Google DeepMind’s AlphaFold in some benchmarks and will be open-source for non-commercial use, potentially accelerating the development of new treatments for diseases.

• A study found that AI-generated research ideas in natural language processing (NLP) are judged as more novel but slightly less feasible than those from human experts. This suggests that AI can contribute significantly to innovation and scientific discovery, despite current challenges with self-evaluation and diversity in output.

• Waymo's new crash data suggests that its driverless cars are significantly safer than human-driven vehicles, with fewer injury-causing crashes and airbag deployments per mile driven. As Waymo scales its robotaxi service to 100,000 rides per week, this data supports the argument that AI-driven cars could greatly reduce road accidents and enhance road safety.

• Google launched Audio Overviews, a new feature in NotebookLM that turns notes, PDFs, Google Docs, and more into AI-generated audio discussions between virtual agents. Leveraging Gemini 1.5’s multimodal capabilities, this tool can process up to 25 million words, offering a new way for auditory learners to engage with academic papers, ebooks, and presentations.

• Adobe previewed its Firefly AI Video Model, set to release by year-end, which includes tools like Text to Video, Image to Video, and Generative Extend to enhance and create video content from text or images. This feature aims to revolutionize video editing by enabling users to modify camera angles, extend scenes, and generate new footage instantly.

• French AI startup Mistral released Pixtral 12B, its first multimodal model that processes both images and text, available for free under an Apache 2.0 license. Despite being just a year old, Mistral is positioning itself as Europe's challenger to OpenAI, advancing rapidly with significant funding and innovative open models.

Quickfire News

• OpenAI reportedly considered high-priced subscriptions: up to $2,000 per month for advanced AI models like Strawberry and Orion.

• Tesla announced plans to launch AI-powered Full Self-Driving in China and Europe: target is early 2025, pending regulatory approval.

• Dell Technologies reported surging AI server sales: CEO Michael Dell dismissed concerns about an AI spending slowdown.

• The US, EU, and UK signed the world’s first legally binding international AI treaty: emphasizes human rights and accountability in AI regulation.

• Microsoft announced its “Microsoft 365 Copilot: Wave 2” event for September 16: promises to reveal next-phase developments for its AI assistant.

• Ada leveraged GPT-4 to double customer service resolution rates: aims for 100% automated query resolution in the near future.

• LMSYS opened beta testing signups for Copilot Arena: a new AI assistant comparison platform for tools like Cursor AI and Copilot.

• OpenAI reported over 1 million paying users across its business products: ChatGPT Enterprise, Team, and Edu within their first year of launch.

• Google Photos upgraded search with natural language queries: launched “Ask Photos,” an AI-powered conversational search feature for US users.

• X permanently stopped Grok AI from using EU citizens’ tweets: following legal action by Ireland’s Data Protection Commissioner.

• Qualcomm CEO revealed a partnership with Samsung and Google: developing mixed reality smart glasses as a companion device for smartphones.

• DeepSeek released version 2.5: a 238B-parameter model with enhanced chat, coding, and 128k context length capabilities.

• OpenAI clarified that “GPT-Next” is not a new model: it is a placeholder term to illustrate potential future AI models.

• YouTube is developing AI detection tools for synthetic music and faces: also adding creator controls for AI model training to protect content creators.

• Audible announced plans to create AI voice replicas of select US-based audiobook narrators: includes a royalty-sharing model for AI-narrated books.

• Oscar-nominated studio XTR acquired AI animation studio Late Night Labs: launched Asteria Film Co. to produce AI-powered animations.

• BP signed a five-year deal with Palantir: aims to expand AI use in its operations and speed up engineer decision-making with LLMs.

• Roche expanded its digital pathology open environment: integrated over 20 AI algorithms from eight new collaborators to enhance cancer diagnostics and research.

• Researchers introduced Robot Utility Models (RUMs): a zero-shot method for training robot policies to perform new tasks in unfamiliar environments.

• Salesforce introduced Industries AI: embeds over 100 industry-specific AI capabilities across 15 industry clouds to address unique challenges.

• Glean raised over $260 million in new funding at a $4.6 billion valuation: co-led by Altimeter and DST Global, to expand its Work AI platform.

• Anthropic unveiled Workspaces in its API Console: allows developers to manage multiple Claude deployments with controls for spend limits, rate limits, and access management.

• UBS developed an AI tool capable of analyzing 300,000 companies in 20 seconds: designed to generate M&A ideas and identify potential buyers.

• The Australian government proposed new legal requirements for AI in high-risk settings: also introduced a Voluntary AI Safety Standard for businesses.

• OpenAI is reportedly in talks to raise $6.5 billion, with an additional $5 billion credit line: would value the company at $150 billion, up from its previous $86 billion valuation.

• Meta is reportedly finalizing a new AI supercomputing cluster: over 100,000 Nvidia H100 chips to train its next Llama 4 language model.

• Canada's Defence Minister announced plans to incorporate AI into its military by 2030: aims to improve the work of military personnel without replacing them.

• Kai-Fu Lee stated that Chinese AI models lag behind U.S. by 6 to 9 months: predicts faster proliferation of Chinese AI apps compared to the U.S. by early next year.

• Taylor Swift endorsed Kamala Harris for president: partly in response to AI-generated deepfakes falsely showing her supporting Trump.

• Oracle unveiled over 50 AI agents for its Fusion Cloud Applications Suite: designed to automate business processes across functions like capital management, supply chain, and customer experience.

Closing Thoughts

That’s it for us this week. And what a week this was ! The autonomous driving news would have been the theme of this edition if not for Strawberry stealing the spotlight.

If you find any value from this newsletter : make sure to send the link below to your colleagues and friends :

https://www.blacklynxbrief.com/subscribe