From dismantling old policies to lifting major export restrictions, a sweeping AI plan to put the US back on top has been unveiled - but is it enough?
Here are our main key takeaways:
- OpenAI CEO Sam Altman announces a $30 billion yearly deal with Oracle to bring 4.5 GW of compute to Stargate
- Intel continues to scale back its operations, canceling major assembly and chip projects in Germany and Poland
- Proton debuts Lumo, the company's take on a real privacy-first AI chatbot
Join us at AI Tangle as we untangle this week's happenings in AI!
|
The Trump administration has launched the "AI Action Plan," an aggressive plan aimed at reasserting the US as the leader in global AI. This includes a 90-point blueprint focused on deregulation, expanded AI exports, and rapid infrastructure development, along with three executive orders that seek to dismantle many Biden-era policies, fast-track data center construction, and support tech giants in selling "full-stack AI" packages to allies. Faced with yet another technological arms race with China, the US is upping the ante as it looks to remove restrictions on chip sales and calls for a single federal standard.
But there's more to it than trade deals and asserting dominance
At the same time, the administration is also moving to forbid "woke AI" from federal use, issuing a separate order that bans procurement of models "influenced by diversity, equity, and inclusion (DEI) principles." The order mandates that federal use of AI should prioritize "objectivity and historical accuracy," rejecting what it calls ideologically influenced outputs. However, although it is steering clear of regulating AI in the private sector, public reception to the move has been divisive, with some calling the administration's take on "ideological neutrality" one-sided and biased in and of itself.
Regardless of stance, the AI Action Plan is a major sweeping initiative and overhaul of the US' AI policy - how it holds up against China and the wider world remains to be seen.
|
🍯 A Sweet Amazon Deal- Amazon acquires AI wearable firm Bee in a likely push to develop its own line of personal AI devices.
❌ The Line Drawn at AI - US management consulting giant McKinsey no longer allows its China business unit to work on generative AI projects.
🖥️ Even More AI Chips - After a record Q2 result, South Korean chipmaker SK Hynix has announced even more investments in AI to meet demand.
👗 A Virtual Try-On - A new update to Google's online shopping experience brings a virtual try-on mode to let users see how clothes look on them.
|
Just a day after reports claimed that Project Stargate was struggling to get off the ground, OpenAI CEO Sam Altman announced a massive deal with Oracle to add an additional 4.5 gigawatts of data center capacity for the $500 billion AI project. The construction of the Stargate I facility is reportedly already underway, with Oracle having begun deliveries of Nvidia GB200 GPU racks to it since last month. Together with SoftBank, OpenAI said in a statement that they are committed to "delivering 10 GW of new compute capacity through Stargate" and creating thousands of specialized jobs.
Intel is further scaling back its global manufacturing footprint under new CEO Lip-Bu Tan, canceling projects in Germany and Poland and consolidating testing to Vietnam and Malaysia. Meanwhile, the $28 billion Ohio chip factory faces another delay after it was already delayed in February. Tan, who took over in March, has been aggressively cutting costs and inefficiencies, including a 15% workforce reduction and eliminating half of management layers as he aims to make Intel refocus on its "core business." Intel now plans to match capacity growth to demand, moving away from past overinvestment.
Proton, known for its Proton Mail encrypted email service, has released an AI chatbot called Lumo, with an emphasis on user privacy. Much like many chatbots, Lumo can draft emails, generate code, summarize documents, etc., and is powered by open-source models from Mistral, Nvidia, and the Allen Institute for AI hosted on Proton's European servers. Lumo protects user data by saving it locally and using "zero-access" and TLS encryption depending on function, preventing anyone external, including Proton, from accessing any information. Lumo is available for free, but a $12.99/mo plan is available for unlimited queries, extended history, and larger file uploads.
A team at Google DeepMind has developed an AI tool, Aeneas, that helps interpret damaged Roman inscriptions and identify when and where they were created. Aeneas was trained on 200,000 Latin texts and has already helped historians link fragments across centuries and regions, giving estimates within 13 years and associating texts with 62 Roman provinces. Early tests show the AI's contextual insights have helped researchers in 90% of cases, with some believing it can make historical research more accessible, something previously only limited to scholars with vast libraries or specialist knowledge.
Walmart has begun rolling out what it calls AI "super agents" to improve customer experience and streamline its operations, with an additional "Associate" in the works. The agents will be tailored for shoppers, employees, suppliers, and developers, and their aim is to centralize and simplify interactions through minimal human input. One of its existing shopping agents, Sparky, is due for one such "super agent" upgrade. With Walmart hoping for e-commerce to account for 50% of sales in five years, a bet on AI could also fuel its aim to take a bite out of online giant Amazon.
|
Kraftul - Kraftul is the copilot for product teams that helps businesses' get insights into what makes their products work and analyzes support tickets, calls, and reviews to more easily understand user feedback.
Mendel - Mendel helps developer teams automate code reviews, optimize workflows, track team performance and enhance deployment efficiency with the help of purpose-built AI.
Rapport - Rapport helps you build rapport with your users by making it easy to deploy virtual interactive personalities, be it for corporate training, entertainment, marketing, or education - in any language.
|
Why Does Thinking Longer Make AI Dumber? (7-min read)
A new study by researchers at Anthropic recently took a deep dive into why longer thinking sometimes makes AI models perform not just not better, but occasionally even significantly worse. What exactly causes this phenomenon, and what does it mean for AI in enterprise?
|
| What did you think of this edition? Let us know - your review helps us improve! |
|
|
|
|
|