Digimasters Shorts - Hugging Face Warns LLM Bubble Burst, Cloudflare Internet Crash Chaos, Google Unleashes Gemini 3 AI, Microsoft’s Risky Windows AI Agents, TikTok's AI Content Control Controversy

Digimasters Shorts is your quick hit of the latest in the digital world. Hosted by Adam Nagus and Carly Wilson, this podcast delivers concise insights on cutting-edge AI developments, industry trends, and web infrastructure challenges. From Google's latest GPT rival Gemini 3 and Microsoft's evolving Windows AI features to TikTok’s AI video revolution and major internet outages, Digimasters Shorts keeps you informed on how technology is shaping our future—one short episode at a time. Tune in for expert analysis and stay ahead in the fast-paced realm of digital innovation.

Transcript

Adam N2: 0:05

Welcome to Digimasters Shorts, we are your hosts Adam Nagus

Carly W: 0:09

and Carly Wilson delivering the latest scoop from the digital realm. Hugging Face co-founder and C.E.O Clem Delangue says the AI industry is not in a bubble but rather experiencing an"L.L.M bubble" focused on large language models like those behind Chat G.P.T. He warns that this L.L.M bubble might burst as early as next year but insists that A.I's broader future remains strong. According to Delangue, L.L.Ms represent only a subset of AI applications, which also include biology, chemistry, image, audio, and video technologies. He argues that the current hype and investment are overly concentrated on building one massive model to solve all problems. Delangue predicts a shift toward smaller, specialized models that will be more efficient, cost-effective, and tailored for specific enterprise needs. He acknowledges that a bursting L.L.M bubble could affect Hugging Face but believes the diverse AI ecosystem will absorb the impact. The company holds half of its$400 million funding in reserve, adopting a cautious and sustainable approach unlike many competitors burning billions. Delangue critiques the current rush and panic in the industry, drawing on his 15 years of AI experience to advocate for long-term planning. He emphasizes Hugging Face’s commitment to building a lasting, impactful company rather than chasing short-term gains. Meanwhile, StrictlyVC continues its 2025 series with exclusive events featuring top venture capitalists and AI builders.

Adam N2: 1:34

Cloudflare experienced its worst outage since 2019 due to a bad query setup in its Bot Management system, which crashed large parts of the internet for several hours. The system is designed to control automated crawlers scanning websites, affecting about 20 percent of the web running through Cloudflare's network. The outage disrupted major platforms including X, Chat G.P.T, and Downdetector. The root cause was a change in the database's permission system, not a cyberattack or generative AI technology as initially suspected. A faulty query caused the ClickHouse database to generate numerous duplicate feature rows, bloating the configuration file beyond memory limits. This overwhelmed Cloudflare's core proxy system, which processes traffic dependent on the bots module. Consequently, legitimate traffic was wrongly blocked while users not relying on bot scores remained unaffected. Cloudflare's bot controls are intended to mitigate automated scraping, including AI crawlers disregarding no-crawl directives. The company has outlined four plans to prevent similar issues in the future but warns that internet centralization may still lead to outages. This incident highlights the vulnerabilities in web infrastructure reliant on centralized services. Google has unveiled Gemini 3, the latest and most advanced version of its flagship large language model, touting it as their smartest model ever with impressive benchmark results. Gemini 3 Pro is now available in preview across various Google products, including Search, while the enhanced reasoning version, Gemini 3 Deep Think, will roll out to select subscribers after safety checks. The model has topped the LMArena leaderboard, surpassing competitors like Grok 4.1, and demonstrated PhD-level reasoning on complex tests. Google also announced Google Antigravity, an AI-powered integrated development environment that assists with coding across multiple platforms, now open for testing on Windows, Mac, and Linux. Developers can access Gemini 3 Pro through Google’s Vertex AI and AI Studio, with the model achieving top marks in coding benchmarks as well. Notably, Gemini 3 is integrated directly into Google Search, making its presence unavoidable for users seeking AI-generated answers. The Gemini app offers a generative interface featuring visual and dynamic views, the latter allowing interactive, website-style outputs. This release intensifies the competition in AI, following recent launches from Open A.I and Elon Musk’s xAI. Google's rapid rollout positions Gemini 3 as a new leader in the AI race, though how long this advantage will last remains to be seen. The AI landscape continues evolving swiftly, with each major player pushing boundaries.

Carly W: 4:23

Microsoft is advancing AI integration in Windows 11 with new generative and"agentic" AI features designed to perform tasks autonomously. A recent Windows Insider build introduced an experimental toggle for"agentic features," enabling Copilot Actions to handle complex tasks like organizing files and scheduling meetings. These AI agents operate with separate user accounts and isolated desktops to protect the system and user workflow. Users must approve data requests, and all agent activities are logged and observable to ensure transparency. Despite these safeguards, Microsoft warns of privacy and security risks, including potential exploits like cross-prompt injection that could lead to data breaches. By default, these features are disabled and can be switched off in settings to give users control over their data. Microsoft’s cautious approach reflects lessons learned from past issues with Windows features that overreached on data collection. The company is also enhancing Copilot by introducing"Mico," an animated assistant to improve user interaction through voice and traditional input. These developments highlight Microsoft's aim to create an AI that acts as an active digital collaborator while balancing security concerns. Users can expect to see these AI capabilities become more prominent in future Windows 11 updates. TikTok has announced it hosts more than 1.3 billion AI-generated videos and is introducing a feature allowing users to adjust how much AI content appears in their feeds. This change, currently being tested, responds to the rise of AI video tools like Open A.I’s Sora and Google’s Veo 3. Users can now manage their preferences under the"AI-generated content" setting, similar to filters for topics like fashion or current affairs. TikTok mandates that creators label realistic AI-made videos and bans harmful deepfakes of public figures or crisis events. Videos created with TikTok's own AI tools or identified by the C2PA initiative will carry an"AI-made" watermark to ensure transparency. The company is also launching a$2 million AI literacy fund to promote responsible AI use through educational content. Despite these initiatives, TikTok plans to cut 439 U.K content moderation jobs, sparking concerns about AI replacing human oversight. TikTok insists that AI aids human moderators by filtering distressing content, reducing their exposure by 76% over the past year. The platform emphasizes a balance between technology and human review to maintain safety. TikTok’s European director of public policy stated the move gives users more control based on their preferences and feedback.

Don: 7:03

Thank you for listening to today's AI and Tech News podcast summary... Please do leave us a comment and for additional feedback, please email us at podcast@digimasters.co.uk You can now follow us on Instagram and Threads by searching for@DigimastersShorts or Search for Digimasters on Linkedin. Be sure to tune in tomorrow and don't forget to follow or subscribe!