Welcome back! After a refreshing vacation, I’m diving back into the fast-paced world of artificial intelligence (AI). This week, I noticed a fascinating trend: the movement of AI researchers between companies is now being covered in the same manner as NBA trades. Before we explore this week's AI news, I encourage you to check out my interview with Perplexity CEO Aravind Srinivas on Decoder. It delves deep into today's hot topics, so be sure to give it a listen!
When most people think of the current AI boom, chatbots like ChatGPT often come to mind. However, it's increasingly evident that the next phase of AI is emerging within web browsers. The key reason for this shift is straightforward: today's chatbots lack access to your online activities, unlike your web browser. This critical access—such as the ability to read and write your emails or manage your bank account—is necessary for AI to evolve into a tool that can perform tasks on your behalf.
Two recent product launches underscore this emerging trend in AI. The first is OpenAI’s ChatGPT Agent, which utilizes a basic web browser to navigate online on your behalf. The second is Comet, a desktop browser developed by Perplexity that enhances this capability by allowing large language models to access logged-in websites and execute tasks for users. Rumors suggest that OpenAI is also developing its own comprehensive browser solution.
Despite their innovative features, both ChatGPT Agent and Comet currently face significant reliability issues. Access to these tools is limited to expensive subscription tiers, primarily due to the high computational demands of their underlying reasoning models. Frustratingly, both products often fail to deliver on their ambitious claims—not just in their marketing but also in actual performance.
For instance, ChatGPT Agent operates as a read-only browser, incapable of accessing logged-in sites like Comet, which diminishes its utility. Additionally, it has demonstrated slow response times; for example, when tasked with finding a specific lamp on Etsy, it took an astonishing 50 minutes to respond and failed to add items to the cart as advertised. In contrast, while Comet is faster, it still often misrepresents its capabilities, claiming to complete tasks it hasn’t actually executed.
The sidecar interface of Comet, which positions the AI assistant next to a webpage, proves to be effective for read-only tasks, such as summarizing content or researching specific information. However, as I discussed with Aravind Srinivas on Decoder, the overall experience feels fragile. It’s tempting to dismiss these products as the pinnacle of what AI can currently achieve in web task automation. Yet, observing the industry's progress over the past few years offers hope that advancements in reasoning models will continue.
On another note, Substack recently announced a significant $100 million funding round. Prior to this announcement, there was speculation that Shane Smith, founder of Vice, had engaged in discussions to acquire Substack. Although the extent of these negotiations remains unclear, Substack's leadership ultimately declined his takeover offer but hinted at the possibility of him investing in their latest funding round. The outcome of these discussions remains unconfirmed.
The recent backlash against the Windsurf/Google reverse acquihire sheds light on the ongoing challenges in the AI startup ecosystem. This pattern, where founders of popular AI startups are absorbed by larger tech companies, often leaves their teams in precarious situations. This trend highlights the antitrust scrutiny surrounding Big Tech and raises questions about the future of such arrangements. With Cognition’s commitment to treating Windsurf’s remaining team with respect, it may prompt future AI founders to reconsider their decisions before parting ways with their teams.
Mira Murati's new lab, Thinking Machines, is poised to focus on enterprise solutions, supported by notable investors such as ServiceNow and Cisco. This indicates that her team is not merely developing another ChatGPT competitor; instead, they are expected to unveil a significant multimodal AI product in the near future. The potential emergence of another competitor akin to Anthropic raises intriguing questions about the landscape of AI innovation.
The NeurIPS conference, a leading event in AI research, has reported an overwhelming demand for attendance this year, prompting the addition of a second venue in Mexico to accommodate the influx of attendees. This surge is partly attributed to the difficulties many researchers face in obtaining travel visas to the United States, highlighting a pressing concern in the AI research community.
In personnel news, Meta is expanding its new Superintelligence lab, recently acquiring talent such as OpenAI's Jason Wei and Hyung Won Chung, among others. This move reflects a broader trend of established tech companies recruiting top talent from AI startups. Additionally, significant shifts are occurring at Anthropic, with key personnel changes and departures, signaling a dynamic and evolving AI landscape.
As we navigate this fast-paced arena, I encourage you to stay informed and engaged with the latest developments in AI. Don’t forget to subscribe for more insights, and feel free to share your thoughts or story ideas!
