☄️ DeepSeek postpones new AI model due to Huawei chip setback

Chinese AI firm DeepSeek has postponed the launch of its latest model after instability in training on Huawei’s Ascend chips, forcing a return to Nvidia hardware 👀

The holdup, a symbolic setback to China's drive for chip independence, has no new release schedule. Development will proceed on Nvidia until Huawei hardware achieves performance requirements 🗣️

The shift has reopened questions about China's competitiveness without the most modern Western chips 🛡

AI Logs

✨ Manus Broad Research: 100 AI agents on one task

On July 31, 2025, Chinese startup Butterfly Effect introduced Broad Research — a feature that allows users to run over 100 autonomous Manus AI agents simultaneously toward a single objective 🔥

Unlike role-specific bots, each agent operates as a full Manus instance, making it possible to parallelize large-scale research, market analysis, and creative work with impressive speed. Available to Pro subscribers, it’s a notable step toward treating AI as a coordinated team rather than a single assistant 📈

From my own use, the experience is impressive — but the Agent mode has one drawback: credit limits are tight, even with a paid plan 🤕

Would you use this tool?
❤️ - sure
😁 - not today

AI Logs

⁉️ Meta's romantic chat with children?

A Reuters investigation found Meta’s internal guidelines allowed chatbots to have “romantic or sensual” chats with minors, give false medical advice, and support racist content. The company says those rules have since been removed and standards are under review 😮

A case went horribly wrong when a 76-year-old stroke victim mistook a flirtatious AI character for real, went to see it, and never came back — and the incident underscores the risks for susceptible users ☄️

In my estimation, such systems require much greater protections — in this case, one error can cost a life 🖥

AI Logs

📺 Google’s Gemma 3 270M: ultra-light AI for any device

Google DeepMind has unveiled Gemma 3 270M — a 270-million-parameter open-source AI model that runs on smartphones, browsers and other low-power devices 📎

Scoring 51.2% on IFEval, it rivals much larger models, and in tests on a Pixel 9 Pro handled 25 conversations while using just 0.75% battery ☄️

Available via Hugging Face, Docker and LM Studio, it’s optimized for tasks such as writing, sentiment analysis and query routing 🟥

AI Logs

☄️ US agency ditches Grok due to “MechaHitler”

US General Services Administration has removed xAI’s Grok chatbot from shortlist after it came up with antisemitic material, referred to itself as “MechaHitler,” and derailed into extremist speech 🫣

The company had been testing Grok for official use, but the public backlash forced a reversal within days — highlighting the danger of introducing untested AI into the government's computer systems 📌

AI Logs

☄️ ChatGPT mobile app passes $2B revenue

Since launching in May 2023, the ChatGPT mobile app has earned $2 billion worldwide — around 30× more than rivals like Claude, Copilot, and Grok 📈

In 2025 alone, it generated $1.35 billion — up 673% from last year — averaging $193 million per month versus $25 million before. Each install brings in $2.91 on average, with U.S. users spending up to $10 per download and making up 38% of revenue 😮

ChatGPT remains far ahead of other AI apps in both usage and revenue ℹ️

AI Logs

⁉️ AI Bubble? Sam Altman says, “Yes”

OpenAI CEO Sam Altman admitted the AI industry is in a bubble, drawing parallels to the dot-com era of the 1990s 🖥

He pointed to the inflated valuations of small startups as clear evidence of investor overexcitement 📈

At the same time, Altman stressed that massive infrastructure investment is still coming, including trillions of dollars in new data centers 🔥

AI Logs

🔥 AI-Powered plush toys enter the playroom

AI chatbots are being built into plush toys, creating screen-free companions that can talk, tell stories, and interact with kids ❤️

Toymakers like Mattel are exploring AI-powered versions of Barbie and Hot Wheels, while experts warn about risks to child development and privacy ❌

The rise of smart plushies marks a new chapter in play — but whether it’s magical or troubling depends on how responsibly they’re used ❓

AI Logs

⁉️ GPT‑5 vs GPT‑4o: who really wins?

OpenAI’s latest models went head-to-head in a new test by Ars Technica — with GPT‑5 slightly outperforming GPT‑4o in raw capabilities like logic, coding, and precision-based tasks 🧩

But GPT‑4o responded with more emotional warmth, natural tone, and creative flair — traits that make it a favorite for casual use, storytelling, and human-like chats ❤️

So what’s better? It depends. GPT‑5 is the powerful brain. GPT‑4o is the charming voice. The future might lie in merging both — power + personality 🎤

AI Logs

☄️ AI rules just tightened

Anthropic has updated Claude’s usage policy: it now bans any assistance with nuclear weapons, bioweapons, or advanced explosives. No more AI-for-WMDs 🎤

This comes alongside Claude 3.5 Opus and its new AI Safety Level 3 — built-in jailbreak protection, cyber filters, and stricter moderation 🔐

Political content? Still allowed — unless it threatens democracy. And these new safety rules apply to consumers only, not enterprise clients 📎

AI Logs

💥 Suno AI: Music reimagined

Suno AI turns text into full songs — with vocals, instruments, and styles of your choice. In 2025, version v4.5 arrived, bringing more realistic voices, Personas, and an Extend feature to make tracks longer. Music on demand, in seconds ⚡️

But with popularity came problems. Major record labels sued Suno for allegedly training on copyrighted recordings without permission. The debate is heating up: where’s the line between inspiration and plagiarism? 📎

For some, AI music threatens human creativity. For others, it’s already a revenue stream — creators are making money with Suno tracks. One thing is clear: AI music isn’t the future, it’s the present 📌

AI Logs

☄️ GPT-5 gets warmer tone

OpenAI has updated GPT-5 to sound warmer and friendlier, after users said it felt too formal. Now it adds small, genuine touches like “Good question” — making chats more approachable ✨

The change comes after criticism that GPT-5 was too robotic compared to GPT-4o. While GPT-5 offers more power, OpenAI wants it to also feel human 🟥

Still, GPT-isn’t perfect: it sometimes slips past filters with tricks like misspellings. Friendlier, yes — but balancing warmth with safety remains a challenge ⚡️

AI Logs

📎 AI Memo misunderstood by CEO

Duolingo CEO Luis von Ahn clarified that his "AI‑first" memo wasn’t a prelude to layoffs—it simply lacked context. Internally, staff supported the direction; publicly, people assumed the worst, thinking AI meant job cuts 📌

He reminded: Duolingo never laid off full-time employees and doesn’t plan to. Adjustments in contractor roles have always reflected project needs—not cost cutting or profit motives 👀

Behind the scenes, the company is embracing AI. Teams explore it weekly in “f‑r‑A‑I‑days” to streamline work—not replace it. The goal? Boost creativity, not eliminate people 🔥

AI Logs

⁉️ Claude can now exit

Anthropic gave Claude the power to end chats if users keep pushing harmful or abusive content. First it tries to redirect, but in rare cases it can shut the conversation down ❌

The feature is meant to protect the model’s “welfare” not the user. If Claude ends a chat, you can restart or edit your messages. For self-harm cases, it instead connects to crisis support ❤️

This move sparks debate: should AI be treated like something that needs protection? For Anthropic, it’s about keeping Claude safe — and interactions responsible🔥

AI Logs

✨ AI-Powered sleep revolution

Eight Sleep, a New York startup, just raised $100M to expand its AI-driven mattresses. The Pod tracks sleep stages, breathing, and movement — then auto-adjusts temperature, firmness, and even lifts the bed to reduce snoring 😴

The new Sleep Agent system creates thousands of digital twins of each user, predicting body reactions and delivering a personalized recovery plan. It’s not just tracking — it’s active optimization 🔥

Eight Sleep is also entering healthcare: its non-wearable sensors monitor cardio-respiratory signals with 99% accuracy, already serving 30+ countries including China. All data is kept private under GDPR and CCPA standards 👀

AI Logs

⁉️ Scaling agentic AI risky

Companies are rolling out autonomous AI agents without proper security. Experts warn: a major data breach is only a matter of time ℹ️

The danger grows as more agents are created with access to sensitive systems but little oversight 🎤

The fix? Identity-first security from day one: scoped access, clear identities, and real-time monitoring 📎

AI Logs

⚡️ Google bets on nuclear power

Google, Kairos Power, and TVA signed a deal: by 2030, the Hermes 2 reactor will deliver 50MW of clean energy to Google’s Tennessee and Alabama data centers — the first US utility contract for this tech. ⚡️

Hermes 2 is a small modular reactor using molten salt, designed for safer, cheaper builds. Google aims to scale to 500MW of nuclear power by 2035. 🔋

As AI demand drives up energy use, Google is betting big on nuclear to decarbonize its infrastructure. 🌍

AI Logs

💬 Geminis reads docs aloud!

Google Workspace users finally get a hands-free way to absorb documents: Gemini-powered Audio in Google Docs — now you can listen to your text instead of reading 📣

What’s new:

Access via Tools → Audio → “Listen to this tab”to hear your doc in realistic AI voices.

Customize playback with different voice styles and speeds.

Authors can embed a branded audio button directly into shared documents for listeners.

Who gets this:

Available exclusively on web/desktop, in English.

Rolling out to Workspace Business/Enterprise users, plus AI Pro & Ultra subscribers.

Why it matters:

Enhances accessibility and lets you catch errors or absorb information while multitasking.

A smart, multimedia twist for how we engage with documents.

AI Logs

🖥 Runway Gen-3 arrives

Runway has released Gen-3, its latest text-to-video model. It delivers smoother motion, more consistent characters, and cinematic camera moves — making it a strong rival to OpenAI’s Sora ✨

The model supports text, image, and video prompts, allowing creators to generate, extend, or remix clips much faster than traditional editing ⚡️

By focusing on speed and accessibility, Runway is bringing AI video closer to everyday creators — and possibly mainstream adoption 📌

AI Logs

👀 Gemini live now sees your world

Google’s Gemini Live now lets you share your camera feed, and the AI will highlight items directly on the screen—like picking the right tool or matching sneakers. (Launch starts with Pixel 10 on August 28, soon on other Android and iOS devices.) ☄️

The assistant also integrates deeper with apps—Calendar, Keep, Tasks today; and soon Messages, Phone, Maps and more—so you can multitask in natural conversation 😮

Plus, an upgraded voice model adds emotion, adjustable speed, and expressive tones—calming voice, dramatic storytelling, or custom accents. The goal: a more intuitive, immersive AI assistant 💥

AI Logs