Don't Listen to Warren Buffett

Good morning.

Last week at the annual Berkshire Hathaway shareholders meeting, Warren Buffet was asked what he thought about the AI revolution . He talked about the immense dangers of “scamming” and that he’s not ready to invest in AI because of the “many dangers.”.

This, my friends, is an old man who is out of touch. He was long thought of as a visionary, but he’s nearing 100, and while he’s right on the scamming part, his brain cannot comprehend what’s happening.

Then, on the other side, you have Elon Musk, who knows damn well what is happening. He bought Twitter.

Nobody knew why, but this week it became a bit more clear: this is the ultimate dataset for an LLM. And with his new product, "Stories," it will filter out all the noise and bias and give you nice, clean "news.”. The Truth.

You can say what you want about him, but the mission of x.AI is to “advance our understanding of the universe,” and the advent of Stories is part of that understanding.

Exciting times.

Welcome to the Blacklynx Brief.

AI NEWS
Start Folding

• Google DeepMind and Isomorphic Labs have launched AlphaFold 3, an advanced AI model that significantly enhances the prediction of molecular structures, including proteins, DNA, and drug-like interactions. This latest version offers a 50% improvement over traditional methods and extends its capabilities to model a broader range of biological molecules. AlphaFold 3 is accessible for non-commercial use through the AlphaFold Server, enabling researchers worldwide to expedite their studies. This development not only accelerates drug discovery but also broadens our understanding of biological processes, promising substantial impacts on global health and medicine.

• At a recent Q&A at Stanford University, OpenAI CEO Sam Altman expressed his views on the future of AI, describing GPT-4 as "mildly embarrassing" and indicating that future models like GPT-5 will be significantly smarter. Altman emphasized that the financial costs are secondary to achieving advancements in AI, particularly in developing Artificial General Intelligence (AGI). He also highlighted the importance of making advanced AI tools like ChatGPT accessible globally, stressing that it should be free for everyone interested. Additionally, Altman clarified that the recently discussed gpt2-chatbot model is not an iteration of GPT-4.5.

• Microsoft and LinkedIn's latest Work Trend Index Annual Report highlights a significant surge in AI adoption in workplaces, branding 2024 as the 'year AI at work gets real'. The report indicates that the use of generative AI has doubled over the past six months, with 75% of knowledge workers now incorporating AI tools in their tasks. Interestingly, while 78% of these users bring their own AI solutions to work, more than half are hesitant to disclose this usage. Moreover, AI skills are becoming crucial for employment, with two-thirds of leaders stating they would not hire someone lacking AI capabilities.

• New research from Scale AI shows that some AI programs might be remembering specific answers from tests instead of really learning how to solve problems. They made a new test called GSM1k that's a lot like an older test, GSM8k. When they used this new test, they found that many top AI models didn't do as well, which means they might have been memorizing answers rather than truly understanding. This discovery suggests that the way we currently check how smart these AIs are might not be the best, and we might need better ways to test them as they get more advanced.

• Elon Musk has introduced a plan to use his AI chatbot Grok on the X to create summaries of news events and user reactions. Grok will generate these summaries, called "Stories," by analyzing posts on X and updating them with new data as it comes in. Initially, this feature is only available to X's premium subscribers, with plans for a wider release in the future. This feature will allow “news” to emerge from thousands of people talking about a subject, and it’s something that was desperately needed to find truth in what is usually a web of lies.

• While on the subject of Musk, Tesla released a video of their robot Optimus Prime casually sorting batteries in a box in one of the Tesla gigafactories. Cathie Wood, the famous investor, believes that Tesla will become the most important company the world will have ever known because of the societal impact they will create using these android bots and especially the product they will unveil in August : the Robotaxi.

• Microsoft has released a top-secret AI Service geared towards intelligence agencies. It’s uses a closed off GPT-4 model allowing intelligence agencies to use AI to wade through sensitvive and classified documents.

• Microsoft is working on an advanced AI model named MAI-1, spearheaded by former Google DeepMind leader Mustafa Suleyman. The model, featuring a substantial 500 billion parameters, aims to rival leading AI technologies from companies like OpenAI and Google. Microsoft has an extensive partnership with OpenAI that might become strained as they are openly developing a ‘ChatGPT’-killer.

• Talking about OpenAI, they have formed a partnership with Stack Overflow to improve its AI models, particularly for programming-related tasks. By integrating Stack Overflow's extensive archive of community-verified programming solutions, OpenAI aims to enhance ChatGPT's ability to provide accurate coding answers.

• Recent studies from German universities show that both new and seasoned teachers find it challenging to distinguish between texts written by students and those generated by AI. In tests, pre-service and experienced teachers identified AI-generated texts with accuracies of only about 45% and 38% respectively, often overestimating their ability to detect such texts. These findings suggest that current methods and tools for identifying AI-generated content are inadequate, hinting at a potential shift towards integrating AI use in education rather than penalizing it.

• Apple has introduced a new range of iPads with a custom M4 chip during an event in Cupertino, highlighting the chip's advanced AI capabilities. The upgraded iPad Pro, equipped with the M4 chip, boasts a 4x performance increase over previous models, supporting advanced AI tasks with its capability of 38 trillion operations per second. Alongside hardware advancements, Apple also unveiled new AI-powered features such as improved document scanning and enhanced editing tools for video, images, and music.

• Recently, the AI landscape has seen the emergence of two new models named 'im-a-good-gpt2-chatbot' and 'im-also-a-good-gpt2-chatbot', which appear to be advanced versions of the previously mysterious 'gpt-2 chatbot'. These models, accessible only via the Chatbot Arena's battle mode, have sparked intrigue due to their seemingly superior capabilities and lack of official documentation. Sam Altman's recent tweet about one of these models has further fueled speculation that they may be linked to OpenAI (edit : now confirmed)

• OpenAI has postponed its upcoming product showcase, initially set for Thursday, to the following Monday, where it is expected to unveil a new AI-based search product. This move positions OpenAI as a potential competitor to Google and Perplexity in the search arena. The event's delay comes amidst growing competition, notably from Microsoft, which is developing its own AI model, MAI-1.

Quickfire News

• OpenAI's newly discovered data logs suggest a possible launch of an AI-driven search engine on May 9th, potentially rivaling Google.

• Audible now features over 40.000 audiobooks created with the company’s AI “virtual voice” tool. Now authors can turn their writing into audiobooks without investing in voice acting talent. RIP voice acting.

• A high-ranking U.S. official has urged China and Russia to ensure that nuclear weapons control remains in human hands, rather than being automated by AI.

• Microsoft has revised the service terms for its Azure OpenAI service, barring U.S. police departments from employing AI for facial recognition tasks.

• An innovative AI algorithm has revealed over 27,000 previously undetected asteroids in existing telescope data, improving the surveillance of near-Earth objects.

• Abound raised $800M for its AI-powered lending platform that analyzes bank transaction data for personalized loan assessments.

• U.S. President Joe Biden announced a new $3.3B investment by Microsoft to build an AI data center in Wisconsin.

Closing Thoughts

That’s it for us this week.

Have a nice weekend.

If you want a version of this newsletter with pictures etc. Be sure to subscribe below:

https://www.blacklynxbrief.com/subscribe