And we are back with a new Weekly dAIgest, and it has been a full week.
This week in bullet points:
We'll walk you through the top stories.
Moving over to politics, the white house invited the major figureheads of the AI world to discuss the recent advancements. The topic at hand: safety, building upon the ongoing debate. The end result was framed as “morally binding” with no proper law framework being introduced as of yet. Regulators were quick to point out that “self-regulation” is an interesting choice here, given that regulatory guidelines make it much easier to actually enforce safety. With the current administration however, it seems likely that further rules are put in the shelf for now. Alongside the accord, President Trump signed an executive order directing federal agencies to officially adopt the term 'Super Intelligence' (SI). A cool term is you ask us, albeit a tad bit cliché.
Sticking to developments in the white house, the American government released a chatbot called America, specifically for government enquiries. The chatbot runs on Grok and Gemini. It works like every conventional LLM, but strictly adheres to citing over 29,000 government websites. So strictly in fact, that it angered the current administration by citing, in their view, incorrect information. These contradictions were hot-fixed a few days after release, and now the bot should be in line with the administration views. Adding to the launch buzz, users quickly discovered a viral Easter egg: typing 'play Minecraft' prompts the bot to recite a government-themed rewrite of the game's End Poem. We couldn't get it to work today though :(
Currently, functionality is quite limited but America will be expanded later on. As it stands, phase 2 is planned in 2027, but there are no details yet what it entails. America is the first of its kind, with general adoption amongst governments on the globe being quite limited. Reasons are usually due to the fact that government interactions adhere to strict rules and as such are not easily handled by AI. On the other hand, America caters to the increasingly complex web of administration websites, and significantly improves user experience in government affairs.
This week Google announced its highly anticipated answer to the frontier models: Gemini 4 Argon. Although it not accessible for the majority of the user base, being in early access for ultra users. Performance is looking solid on the self-reported benchmarks, but we need to wait to get our hands on it before we can nail it down. In general, Google has slowed its release pace significantly compared to the top AI companies, with many in the space critiquing this strategy. We’d expect however that this announcement will lead to a more general rollout and perhaps even an update of Gemini 3.1 pro (if you know you know).
Given the increasing security breach incidents with AI agents, the U.S. Federal Trade Commission (FTC) opened an investigation into OpenAI and Anthropic. The agency announced a formal investigation, but remained unclear of the exact details. They did highlight concerns about consumer safety in regards to the recent incidents, those incidents led to OpenAI halting training on their most powerful unreleased models. The statement came with a report, detailing OpenAI's view on the Huggin Face breaches.
The FTC now pushing harder on security does make sense in light of the increasing worries on the subject. Many fear that AI agents autonomously gain access to personal information and as such more scrutiny is warranted. The actual results of the investigations will probably take some time, but it will be interesting to see what they think.
Last week we highlighted impressive financials on Deepseek and this week saw similar figures moving around. Starting off, Nvidia's board authorized a record $150 billion increase to its share buyback program, lifting its remaining capacity to $235 billion through fiscal 2028. Reason for buying back shares is further investments in the booming AI market, which of course includes Nvidia itself. AMD also made strides, acquiring World Labs for $8.2B. World Labs is an AI model and research lab, under leadership of Dr. Fei-Fei Li. The acquisition strengthens AMD's position in the AI market, hoping to follow Nvidia with providing hardware.
All in all, these investments hope to strengthen the market, but caution remains. The current fundamentals show eerie similarity to the dot.com bubble, with many fearing an eventual collapse if AI companies can't follow through on promises. Maybe it is time for the Big Short 2?
We are seeing a big shift in how AI is approached, especially on the regulatory front. It will be interesting to see the coming AI developments, but one thing remains sure: we are here to catch you up to speed.