Saturday Links: GPT-5 weeks and months, app builders, European LLMs

Here are this week's AI links:

• App builders turn app design upside down (Hat tip to Victor Delgado for the link and the framing). There are a myriad of GenAI app creation tools now, like lovable, which Victor mentions. These allow you to use text prompts to create fully functioning web apps (and phone apps). The big change is that it's now easy to get functionality first, test that, and then worry about neatness and design later. That's a huge win when trying to design functionality and processes + get user feedback. The big long-term question, though, is, will we need custom apps at all? At what point will we be using to custom, on-the-fly UIs for tasks as they come up?
• OpenAI Roadmap Update for GPT-4.5 and GPT-5. Sam Altman dropped a roadmap update on X/Twitter this week. There are no dates but a statement that from the next releases, we'll see the naming scheme become more rational. Rather than all the version numbers, you'll be using GPT-5 at different levels of fidelity. This certainly makes sense for a consumer audience that really doesn't want to know about the differences between models. In response to a question on the timeline? "Weeks? Months?" Altman replied, "weeks/months". It's a pretty worthy GPT-like answer.
• Legal AI startup Harvey lands fresh $300 million in Sequoia-led round. Vertical AI is still flying. Harvey is one of the original breakout vertical applications of LLMs and has passed $50m in annual recurring revenue. That's impressive growth in just a few years. There are plenty of competitors in the space so it will be interesting to see if there end up being a few big winners or a lot of fragmentation. My guess would be that it will be a few big winners. Although it is now cheaper to catch up technically (you probably don't need so much investment), brand strength and breadth of offering is likely to play an important role in adoption.
• European AI alliance unveils LLM alternative to Silicon Valley and DeepSeek. Diverse approaches to tech and collaborations are always a good thing. But... 52M of funding? Even Deepseek had way more than this just in hardware (their historical hardware purchases are estimated at $500M on the high end). I just hope there are spinoff benefits from the constraints, or at the very least, they can use a base model like Aleph to start. Otherwise, there is no hope of producing any effective here.
• UK and US refuse to sign international AI declaration. Another big news story this week was the UK not signing the Paris AI declaration, possibly in sympathy with the US position. The declaration pledges an "open", "inclusive", and "ethical" approach to AI. The actual declaration is here, and there is also a military-focused statement here. The main declaration looks fairly general, and it seems hard to disagree with anything it says. The military declaration seems about as much of a no-brainer to sign as you can get (e.g., "Underlining that responsibility and accountability can never be transferred to machines," etc.). US Vice President JD-Vance's stated reason for non-signature was that restrictions would stifle innovation, which at least makes motivations clear. The UK government line was centered around the text not going far enough: "“We agreed with much of the leaders’ declaration and continue to work closely with our international partners. This is reflected in our signing of agreements on sustainability and cybersecurity today at the Paris AI Action Summit,” the spokesperson said. “However, we felt the declaration didn’t provide enough practical clarity on global governance, nor sufficiently address harder questions around national security and the challenge AI poses to it.” (Quoting from the Guardian). Seemingly trending very strongly towards Yes Minister.

Wishing you a great weekend.