OpenAI on the Verge of Bankruptcy

Good morning.

As predicted in a previous edition of this newsletter the end of 2024 and 2025 will be a period in which the field of artificial intelligence will be going through the “trough of disillusionment”.

The AI sceptics will have seem to have won.

AI will have seem to have undelivered on its promise.

This sentiment is already playing out on the stock market. NVIDIA - the chip maker that is leading the way in delivering the processing power for the AI revolution saw its stock diminish by 10% on one day this week shedding 3B$ in market value on one day.

Investors are getting nervous. Enormous amounts of money has been pumped into AI-companies but where is the profit ?

OpenAI will be posting a 5B$ operational loss by the end of 2024. Training AI models is incredibly expensive and a massive headache as you’re being sued from all sides for copyright infringement.

In 2023 OpenAI spent 7 Billion dollars on training its models and 1.2 Billion on employee wages. They’re on the verge of bankruptcy.

ChatGPT costs 700.000$ per day to operate.

In the light of this we can better understand all the OpenAI drama over the past year, extensively documented in this newsletter.

The last thing you want as a CEO trying to turn your company profitable is someone that wants to halt development because of safety concerns. Even if they’re warranted.

What Sam Altman basically is doing is gambling. 10 Billion dollars on red please !

He’s gambling on OpenAI reaching AGI or Artificial General Intelligence. Once OpenAI achieves this they will go from a company hanging on the verge to exponentially the most valuable company this planet will ever see.

OpenAI will run the world.

If OpenAI were to develop AGI (Artificial General Intelligence), it could license or sell the technology to governments and corporations.

When this happens - our world would change drastically. And I’m not sure that it will be for the better.

Welcome to the Blacklynx Brief!

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 10€ Amazon gift card.

Scroll to the bottom of the newsletter for details !

AI News
OpenAI’s Memory Feature Live in Europe

• OpenAI’s “memory” feature became available in Europe. We tried playing around with it and it REFUSES to store personal data in its memory. On YouTube you can see US Citizens point ChatGPT towards their Linkedin profile and absorb the info. That’s the EU for you

  

  
  
  

• Alibaba unveiled Qwen2-VL, a vision-language AI model that outperforms GPT-4o in benchmarks, particularly in document comprehension and multilingual text-image understanding. The model excels in tasks like college-level problem-solving and supports multiple languages, signaling Alibaba's strong entry into the state-of-the-art AI arena.

• OpenAI and Anthropic signed an agreement with the U.S. Artificial Intelligence Safety Institute, granting the government access to test their AI models before public release. This move, alongside increasing regulatory scrutiny, marks a significant shift in how AI models are developed and regulated, potentially reshaping the global AI landscape.

• Magic introduced LTM-2-mini, an AI model capable of processing 100 million tokens of context, vastly surpassing current models. Partnering with Google Cloud, Magic aims to build advanced AI supercomputers, positioning the company at the forefront of creating AI assistants with extensive memory and recall capabilities.

• 1X Technologies unveiled NEO Beta, an AI-powered humanoid robot designed for home assistance, with human-like movement and interaction capabilities. Standing 5'5" tall and weighing 66 pounds, NEO Beta can perform tasks ranging from household chores to providing support for those with mobility challenges. This advancement positions 1X Technologies as a serious competitor in the growing humanoid robotics market.

• Amazon's upcoming “Remarkable” version of Alexa, set for release next month, will use Anthropic’s Claude AI models instead of Amazon’s in-house technology, following performance issues with the latter. The new AI-powered Alexa aims to offer enhanced conversations and home automation, with a subscription fee of $5 to $10 monthly, while the classic version remains free.

• Nous Research introduced DisTrO, a new training method for AI models that drastically reduces the need for super-fast internet connections, potentially enabling AI training on regular home computers. DisTrO cuts data-sharing needs by up to 10,000x during fine-tuning, making decentralized AI training more accessible to individuals and smaller organizations.

• Elon Musk's xAI launched "Colossus," the world's most powerful AI cluster, featuring 100,000 Nvidia H100 GPUs, built in just 122 days. xAI plans to double the size to 200,000 GPUs soon, significantly boosting its AI capabilities and pressuring competitors like OpenAI and Google.

• OpenAI is reportedly developing its first in-house AI chips using TSMC's advanced A16 process, potentially partnering with Broadcom, Marvell, or Apple for design expertise. This move could reduce OpenAI's reliance on Nvidia and enhance its AI development efforts.

• Researchers at Profluent Bio used AI to create OpenCRISPR-1, a novel gene-editing tool that matches or surpasses current CRISPR systems in performance and safety. OpenCRISPR-1, designed by Large Language Models and made publicly available, could accelerate the development of personalized treatments for genetic diseases.

• OpenAI Japan’s CEO revealed that ‘GPT-Next,’ expected to be 100 times more powerful than GPT-4, is in development for a potential release by the end of 2024. The model's leap in performance will come from improved architecture and efficiency rather than increased computing resources, with a more advanced version called Orion slated for 2025.

• Altera’s Project Sid has created the first simulation of over 1,000 autonomous AI agents collaborating in a Minecraft world, forming economies, cultures, and governments autonomously. This breakthrough in AI agent collaboration could have wide applications, allowing us to test complex societal solutions in virtual environments before implementing them in real life.

• The U.S. Department of Justice has issued a subpoena to Nvidia, escalating its antitrust investigation into the company’s practices in the AI chip market. The probe centers on concerns that Nvidia may be restricting competition, which could lead to new competitors entering the AI chip space, especially as Nvidia's shares took a record $279 billion hit following the news.

• Safe Superintelligence (SSI), co-founded by former OpenAI chief scientist Ilya Sutskever, raised $1 billion in funding just three months after its launch. The startup, valued at $5 billion, will focus on developing safe AI systems that surpass human intelligence, emphasizing R&D before releasing products, with backing from major investors like Andreessen Horowitz and Sequoia Capital.

• Groq introduced LLaVA v1.5 7B, a multimodal AI model that understands both images and text and runs 4x faster than OpenAI’s GPT-4o. This model, capable of tasks like image captioning and visual inspections, is offered in free preview mode and highlights Groq’s strength in fast processing, opening doors to new AI applications.

• Researchers discovered that making AI systems "re-read" questions twice (RE2) significantly enhances reasoning and problem-solving across tasks like math and symbolic reasoning. This simple human-inspired technique can be applied to various AI models, showing the potential for overlooked methods in improving AI performance.

Quickfire News

• Meta reported significant growth for its Llama AI models: downloads approaching 350 million, with usage increasing 10x since January.

• Nous Research released the Hermes Function Calling V1 dataset: designed for training AI models in function calling and structured output capabilities.

• Nvidia and Apple reportedly discussed joining OpenAI’s funding round with Microsoft: potential valuation of the AI startup could exceed $100 billion.

• California lawmakers approved a bill proposing sweeping AI regulations: includes safety testing requirements and potential legal consequences for harmful AI systems.

• Yale University announced a $150 million investment over 5 years: aimed at supporting AI research, development, and education across the institution.

• Codeium raised $150 million in Series C funding: reached a $1.25 billion valuation and unicorn status less than two years after launch.

• Playground launched a new AI-powered graphic design tool: allows users to create logos, social media designs, t-shirt designs, and more for free.

• Google partnered with Indian startup Salcit Technologies: to develop AI that detects tuberculosis and other diseases from cough sounds.

• Amazon hired Covariant AI robotics experts: licensed their foundation models to advance warehouse automation capabilities.

• Oprah Winfrey announced a primetime ABC special on AI’s societal impact: will feature interviews with tech leaders like Sam Altman and Bill Gates.

• Stanford Medicine researchers developed nuclei.io: an AI tool to help pathologists identify diseased cells more quickly and accurately.

• Coinbase executed its first AI-to-AI cryptocurrency transaction: demonstrated AI agents’ ability to use digital wallets for autonomous trading.

• Tom Hanks warned followers about fraudulent AI-generated ads: these ads used his likeness without consent to promote unauthorized products.

• California Assembly passed AB 3211: requires AI image watermarking, potentially banning most AI image generation in the state.

• Researchers developed an AI tool that models cellular metabolic states: combines different data types to create kinetic models.

• UK’s first “teacherless” classroom will open at David Game College in London: uses AI platforms to provide personalized GCSE instruction.

• Tesla is reportedly planning to reveal its Robotaxi self-driving cars: the event is set for October 10 at California Warner Bros. Studio.

• SportsLine’s AI-powered PickBot released NFL predictions for all Week 1 games: based on each team’s historical data and opponent strengths.

• Anthropic launched a Quickstarts repository: helps developers build applications with their API, starting with a Claude-powered customer support agent app.

• Google announced the release of Function Calling in Google AI Studio: allows users to test model capabilities directly within the user interface.

• Ethereum founder Vitalik Buterin suggested AI and prediction markets: could accelerate X’s Community Notes feature for faster fact-checking.

• The FDA approved Rebrain’s OptimMRI software update: includes a new AI model to help target brain regions for treating Parkinson’s disease.

• Dutch regulators fined Clearview AI over $30 million: due to GDPR violations, including building a biometric database without consent.

• Ligo released an open-source implementation of DeepMind’s AlphaFold3: focused on protein structure prediction.

• A new benchmark of 80+ large language models found Anthropic’s Sonnet 3.5 performed best overall: top models varied significantly by programming language.

• Anthropic launched Claude Enterprise: competes with ChatGPT Enterprise by offering larger context windows and GitHub integration for businesses.

• Google committed $10 million: aims to improve AI literacy among young people in 17 countries, across 13 different languages.

• Sakana AI raised $100 million in Series A funding: announced a collaboration with Nvidia to advance AI research and infrastructure in Japan.

• Salesforce agreed to acquire AI voice agent firm Tenyx: enhances its AI offerings as part of the race for AI talent and technology.

• A Samsung executive reportedly revealed GPT-5 specs: 3.5 trillion parameters and 7000 B100 GPUs for compute, at SEMICON Taiwan.

• Microsoft announced Copilot+ AI features will expand to new laptops: starting in November, broadening AI capabilities beyond ARM-based devices.

• You.com raised $50 million: focuses on AI productivity agents for complex queries, shifting away from general AI search.

Closing Thoughts

That’s it for us this week.

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