Welcome to AI's Eye! With this edition….Scale AI will cut 14% of the workforce after Meta investment and reduce employment for founder Wang...Openai cuts sales of ChatGpt shopping with Hunt for revenue...Former top Google researchers have built a new kind of AI agent. They believe it could be a step towards super intelligence.
This week we are turning our eyes from Vancouver to AI. In Vancouver, I met thousands of PHD-level AI researchers at the International Machine Learning Conference (ICML), one of the top gatherings for AI talents at elite universities, big high-tech labs and AI startups.
I'm honestly humbled to be here. Surrounded by professors, postdocs and industry researchers, he casually drops mathematical evidence and references to thermodynamic metaphors into everyday conversation. There are thousands of posters, papers and presentations. Even if I supply them all to ChatGpt, there are too many posters to be meaningfully absorbed. My brain feels like I'm running at maximum capacity as I try to understand the lectures entitled “Control the underestimation bias in constrained reinforcement learning for safe search” and “Matching individual flows for graph generation.”
Still, there's electricity about being in a room where the future of AI is discussed, defined and possibly redirected. I am very much insisting that today's theory falls from my comfort zone and into the novice mindset, especially in a space where today's theory could become tomorrow's technology. I close the time with ICML and here's what sticks to me.
The AI talent war is in full swing. Meta's extraordinary, ongoing employment has thrown billions of dollars at elite AI researchers seduced by Openai, humanity, Google Deepmind and others, but ICML's story. Some say one London-based researcher is creating a “bubble,” while others see it as an overload. Others expect Meta to be one of Swedish doctoral students to show off herds from various exhibition hall booths, including blue IBM socks. In any case, Big Tech had recruiters working overtime. Private after-hour events for candidates were studded at a venue near the Vancouver Convention Centre.
There is no good place to ask questions. One of the best things about spending time with truly clever people is choosing the brains of truly clever people. It is usually helpful for AI researchers to be extremely kind in explaining their work to curious journalists. In a vast expo hall filled with seemingly endless rows of poster presentations, it is unclear when events will happen to a Stanford professor who is completely satisfied with 20 minutes of chatting about the behavior and ethics guardrails of AI models. How valuable is a one-on-one meeting here? How about machine learning pioneers philosophically waxing the past (this is only 10-5 years ago). Or is a former Google brain researcher patiently explaining the in and out of trans models? valuable.
Scaling up Reinforcement Learning (RL) is all angry. It's not my line – it comes from X's former Tesla AI chief Andrej Karpathy, but RL was all over ICML. (RL, or reinforcement learning, is a training method in which AI learns through trial and error and maximizes some reward. It could be the points in the game, or the number of outputs of the AI model that it receives from a human evaluator.) With more data, more computing, and more difficult tasks, you will get a model that works to infer (hopefully) and more surely follow the instructions, and behave more securely in a messy, real-world enterprise setting.
Many researchers are ready to live their founder's dreams. There is an incredibly young duo of Princeton Building multimodal medical foundation models. Waymo's PHD-level internships using Pokemon Games stress test large language models and AI agents on strategy. A former Google and former Operanny researcher who aims to make a leap in current AI technology. With VCS walking through the hall during the day and offering open bar events at night, this research conference was easy to consider as a breeding ground for current and potential founders.
I'm ready to go back to New Jersey and towed by a bunch of new story ideas and sources. So, this is the rest of the AI news.
Sharon Goldman
Sharon.goldman@fortune.com
@sharongoldman
News ai
Scale AI will cut 14% of the workforce after Meta investment and reduce employment for founder Wang. Just weeks after Meta invested $14.3 billion in scale AI and hired founder Alexandre Wan, the AI startup has fired 200 full-time employees. According to CNBC. In a note to staff on Wednesday, interim CEO Jason Droege, who recently took over Wang, said the company has expanded its “too fast” generative AI efforts and built “over-bureaucracy.” Despite the cut, Droege emphasized that the scale remains well funded and resourced. “These changes will make us more agile and allow us to respond quickly to changing market and customer needs,” he wrote in a note viewed by CNBC. “This structure allows us to better serve today's customers and attract customers who have slowed down their work.”
Openai cuts sales of ChatGpt shopping with Hunt for revenue. According to Financial TimesOpenai plans to cut down on product sales directly created through ChatGpt to bolster its efforts to turn chatbots into an e-commerce platform. Now, ChatGpt is surfaced with a product list with links to external retailers. However, according to multiple sources familiar with the plan, Openai is currently aiming to integrate a complete checkout system. Merchants who meet the order through this system will pay Openai a committee. The company also announced its partnership with Shopify in April as part of a broader push to AI-driven shopping.
Former top Google researchers have built a new kind of AI agent. They believe it could be a step towards super intelligence. Their startup, Reflection, has announced Asimov, an agent who learns how software is built by not only reading the code but also digesting everything around it. Update emails, slack messages, documents, and projects. According to Wiredthe goal is to create more capable AI assistants and teach models how humans actually build things.
AI's good fortune
Elon Musk has released Xai's Grok 4 without a safety report. – Beatrice Nolan
Commentary: Companies that are firing AI staff today will regret it in 5 years – Alexandra Evert
Delta moves towards eliminating AI-favourable set prices that determine how much you personally pay for your ticket – Irina Ivanova
The bewildered CEO of a $14 billion AI company says the secret to success is that your competitors “sleep in that fear” of stealing your ideas – By Preston Fore
AI Calendar
July 22nd-23rd: Fortune Brainstorming AI Singapore. Please apply to attend here.
July 26th-28th: world Artificial Intelligence Meeting (WAIC), Shanghai.
September 8th-10th: Fortune Brainstorm Tech in Park City, Utah. Please apply to attend here.
October 6th-10th: world ai Weekly Amsterdam
December 2-7: Newlips, San Diego
December 8th: Fortune Brainstorming ai San Francisco. Please apply to attend here.
Turn your eyes to AI numbers
75.6%
This is how much startup funding has skyrocketed in the first half of 2025 thanks to the ongoing AI boom. According to new data from Pitchbook, the US startup raised a total of $162.8 billion in the first six months of the year.
That previous surge was driven by ultra-low interest rates during the Covid-19 era, but this time AI is driving momentum. Despite the continued struggle for funding for many venture funds, large investments from big technology and AI native companies are overcharging the market. In the last three months, startups have brought in $69.9 billion.
Notable deals include Openai's $40 billion round, Meta's Scale AI's $14.3 billion stakes, and more than $1 billion in salary increases for safe ultra-intelligence, thinking machinery, Andrill, and grammar. AI accounted for 64.1% of total trading value in the first half and 35.6% of trading volume.
