- AI For All
- Posts
- Breaking Point
In partnership with
Hello readers,
Welcome to the AI For All newsletter! Microsoft and OpenAI are having relationship troubles, OpenAI is trying to ship Orion by December, AI agents are all the rage, and new reports show that GenAI ROI is declining. What’s going on? Let’s find out!
Breaking Point
The Satya gang
A report from The New York Times reveals that Microsoft and OpenAI’s partnership is “showing signs of fraying.” Satya Nadella’s early investment in OpenAI, once heralded as a brilliant 4D chess move, might be shaping up to be the biggest regret of his career. To no one’s surprise, OpenAI is under dire financial pressure and apparently asked Microsoft for more money and compute, which Microsoft resisted.
Hedging its bets, Microsoft acquired a struggling Inflection AI to reduce its dependence on OpenAI’s models. Inflection’s CEO, Mustafa Suleyman, proceeded to fail upwards to become EVP and CEO of Microsoft AI, at which point he apparently yelled at an OpenAI employee for not delivering new models fast enough. To no one’s surprise, Suleyman is a manager, not an engineer. Nowhere is this more obvious than his TED Talk where he sported his Steve Jobs cosplay. Happy Halloween!
There is a clause in OpenAI’s contract with Microsoft that states that if OpenAI builds AGI, Microsoft loses access to its tech. So, if OpenAI ever wanted to cut ties with Microsoft, it could just prematurely claim that AGI has been achieved. After all, OpenAI is intentionally vague about what constitutes AGI. Given this context, should OpenAI ever claim that AGI has been achieved, you should be very skeptical.
o1 was rushed, and Suleyman is breathing down everyone’s neck, so it’s no surprise that OpenAI is trying to release Orion, its new flagship model, by December. Orion is the successor to GPT-4, so why not call it GPT-5? Probably because it’s not worthy of the hype that GPT-5 generated. Orion may be the last gasp of the scaling hypothesis.
In other news, agent is the AI industry’s latest marketing buzzword. Microsoft Copilot Studio now has “agentic capabilities,” Inflection AI is bringing “agentic workflows” to the enterprise, Asana lets you deploy AI agents in critical workflows, Honeywell is building industrial AI agents, Anthropic’s new Claude models can agentically control your computer, and Cisco has an AI agent too, I guess.
So, are AI agents legit? Remember, these are still error-prone LLMs. There is no new technology here. It’s glorified task automation. Why is enterprise software typically ugly and verbose? The business model of many SaaS companies is to lock-in users and then profit as much as possible per captive user. There is no incentive to make the product better if the user can’t leave. There’s a reason prisons aren’t luxury resorts. AI agents provide perhaps the last bit of marketing language available for SaaS companies to upsell users on their largely unprofitable AI offerings.
A few other items to mention: a new report from Appen found that while generative AI adoption is still on the rise, data quality challenges are also rising. Also, enterprise AI deployments and ROI are declining. This tracks with findings by The Wall Street Journal at their CIO Network Summit, where we got encouraging quotes like “90% of generative AI experiments aren’t making it beyond the lab” and “accuracy and reliability is a big problem.” On top of this, at another summit, Marc Andreessen (of all people) suggested that LLM developers may be in “a race to the bottom” (they are).
Lastly, just so you know, Elon Musk’s X is changing its privacy policy to allow third parties to train AI models on your posts unless you opt-out. This is probably a desperately needed revenue stream for X, which has plummeted in value due to Musk’s gross mismanagement and self-defeating provocations. One thing is for certain, training on X data is not a solution to the challenge of data quality.
🔥 Rapid Fire Inferno
Google DeepMind releases SynthID to watermark AI-generated content
Ex-OpenAI researcher says company broke copyright law, makes case
Cohere introduces multimodal Embed 3 AI search model for businesses
Stability AI introduces Stable Diffusion 3.5 Large, Large Turbo, and Medium
IBM introduces open source Granite 3.0 model family for enterprise AI
IBM launches Guardium Data Security Center for AI and quantum threats
Qualcomm and Google collaborate on in-vehicle generative AI assistants
Qualcomm unveils Snapdragon 8 Elite to power on-device generative AI
NVIDIA advances AI in India with Reliance and Tech Mahindra partnerships
Infosys and University of Cambridge establish enterprise AI lab in London
Meta shares new research, models, and datasets including SAM 2.1
Apple rolls out beta AI features and releases Apple Intelligence APIs
White House memo calls for agencies to protect AI from adversaries
FDA publishes views on AI regulation in healthcare and biomedicine
US launches probe into Tesla’s full self driving software after fatal crash
US military applies AI to intel data sets, drudge work, and adds AI officer
UK regulator plans investigation into Alphabet’s investment in Anthropic
Denmark launches sovereign AI supercomputer for scientific challenges
There’s a reason 400,000 professionals read this daily.
Join The AI Report, trusted by 400,000+ professionals at Google, Microsoft, and OpenAI. Get daily insights, tools, and strategies to master practical AI skills that drive results.
📖 What We’re Reading
“After all the hype over AI, the value is hard to find. CEOs have authorized investments, hired talent, and launched pilots—but only 22% of companies have advanced beyond the proof-of-concept stage to generate some value, and only 4% are creating substantial value, according to new BCG research.”