Subscribe
Share
Share
Embed
Daily AI news and research, distilled. UpNext AI breaks down the most important developments in artificial intelligence—from major industry moves to cutting-edge papers.
Welcome to the UpNext AI podcast. It's Thursday, May 7th, 2026, and here's what matters in AI today.\n\nFirst up, DeepSeek could be heading for a roughly 45 billion dollar valuation in its first investment round. TechCrunch, citing reporting from the Financial Times and Bloomberg, says the Chinese AI lab is in talks for financing that would value the company far above an earlier reported 20 billion dollar figure. The key point here is not just the number, but what it says about how fast the market is repricing frontier model challengers.\n\nDeepSeek rose to prominence after launching a large language model that, according to the reporting, trained with a fraction of the compute and cost used by major U.S. labs. Since then, it has stayed in the conversation on reasoning and coding while also keeping open-weight versions available. TechCrunch reports that this would be DeepSeek’s first outside investment round, and that the fundraising could help it compete for talent as rivals try to poach researchers.\n\nSo the broader takeaway is simple: capital is still flowing toward labs that can credibly claim frontier performance, especially if they also offer a different cost structure or strategic position. In DeepSeek’s case, that also means a major China-based contender is being valued more like a central player than a side story. And for now, that 45 billion dollar figure is still reported talks, not a finished deal.\n\nFor our second story, Snap says its 400 million dollar deal with Perplexity has amicably ended. TechCrunch reports that the partnership, announced last November, would have integrated Perplexity’s AI search engine directly into Snapchat. Perplexity was set to pay Snap 400 million dollars in cash and equity over one year, but Snap now says the relationship ended in the first quarter, and its sales guidance assumes no contribution from Perplexity.\n\nWhat makes this worth watching is that it is a very public example of how hard consumer AI distribution deals can be to actually land. On paper, the logic was clear: put conversational search into a massive social app and create a new way for users to ask questions inside Snapchat. In practice, TechCrunch reports that Snap had already said in February that the two sides had not yet agreed on a path to broader rollout, even though the integration had been tested with select users.\n\nThat makes this more than a broken partnership headline. It is also a reminder that distribution, product fit, revenue sharing, and rollout terms are still unsettled across a lot of consumer AI tie-ups. Big announcement value does not always translate into durable deployment.\n\nNow to today’s research pick. Researchers writing in Scientific Reports describe a deep learning framework for tactical analysis in football, and the interesting part is where they focused the system. Instead of trying to solve everything from raw video perception all the way to final analysis, the paper concentrates on structured tracking data and the reasoning layer on top of it.\n\nThe system has two main pieces. One module, called Tactiformer, uses a transformer-based approach to model how players move and coordinate over time, with attention to roles and zones on the pitch. A second module, called StratGaze, looks across segments of play to identify recurring tactical patterns. In plain English, one part tries to understand the motion and relationships between players, and the other tries to recognize bigger strategic motifs that repeat across a match.\n\nAccording to the paper, the framework improved event prediction, trajectory forecasting, and motif clustering on benchmark datasets including SoccerNet and PASS, and the authors also emphasize interpretability through attention maps and tactical timelines. That matters because sports analytics tools are much more useful when coaches and analysts can see why the system thinks a pattern is important, not just get a black-box score.\n\nBottom line: this is a good example of AI being most useful when it narrows in on a specific reasoning problem, rather than trying to automate an entire pipeline end to end.\n\n...Are you building apps with voice? Elevate your app's voice capabilities with ElevenLabs. Their API is a game changer for embedding dynamic, responsive voice interactions in your applications, providing unprecedented realism, flexibility and latency. In fact, you're listening to one of their voices - right - now. If you are a developer looking to elevate user experience with natural voice interfaces, this is your solution. Visit up next dot fm slash eleven to check out their latest offerings. ...\n\nFirst, a new GitHub project called agent-skills-eval is pitching itself as a test runner for Anthropic-style agent skills. The repo says the point is to measure whether a skill actually improves an agent’s performance by comparing runs with and without that skill, then grading the outputs side by side. That is a small but practical sign of where the agent ecosystem is going: less hype around adding capabilities, more pressure to prove they work.\n\nNext, in a piece on Interconnects, Nathan Lambert argues that the term distillation attacks is muddying an important distinction. His point is that distillation itself is a standard and legitimate technique for training smaller or specialized models, while the real concern in current cases is API abuse, jailbreaking, or other attempts to extract information in ways providers did not intend. It is partly a language debate, but language shapes policy, and that makes the distinction matter.\n\nAlso in the mix today, Wired takes aim at the industry’s habit of naming AI features after human mental processes. The example in this piece is Anthropic using terms like dreaming and memories for agent features. The criticism is straightforward: the more human the vocabulary gets, the easier it is to blur what these systems are actually doing under the hood.\n\nAnd finally, The Verge reports that former OpenAI CTO Mira Murati testified in the Musk versus Altman case that she could not trust Sam Altman’s words on safety-related questions around a model deployment process. That is one more sign that the legal fight around OpenAI is continuing to surface internal disputes that go beyond product launches and straight into governance and safety decision-making.\n\nBefore we wrap up, a quick note: this podcast is generated with the assistance of AI and is intended for informational purposes only. All referenced articles, research, and commentary remain the property of their original authors and publishers.\n\nAnd that's your briefing for today. Full source links are in the episode notes, and we'll be back tomorrow with what's up next!