🔁 AI Society 9.1.25 — Meta’s partner play, GPT-5 prompt peek, teen-safe bots, AI browser wars & DeepSeek’s chip pivot

Source: ChatGPT

· Meta’s next move: weighing Google/OpenAI partnerships while Llama-5 bakes; plus a creative boost via Midjourneyand a fresh ARC-AGI snapshot.

· Under the hood of GPT-5: a decoded system-prompt “peek” fuels debate about alignment, capability — and vibe.

· Safety first for teens: Meta tightens guardrails and filters access to risky AI characters/topics.

· AI browser wars: Chrome and OpenAI race to make the browser your AI control center.

· Hardware geopolitics: DeepSeek adds support for Chinese chipsets (Huawei/Cambricon), reshaping the training/inference mix.

· Also worth a look: MemDec (a third path beyond retraining/RAG), top open-source AI UIs (Ollama, LM Studio, Open WebUI), Lightning AI’s multi-cloud GPU marketplace, and why brands must disclose AI use.

— — — — — —

To lead, an update on Meta. ‘The Information’ reports on the company’s plans to potentially partner with either Google or OpenAI to improve its Meta AI chatbot capabilities. Sounds a bit like Apple? This as a potential stopgap while ‘Llama 5’ is baking. Note Meta’s recently announced partnership with Midjourney for image and video creation. For comparison, the latest leaderboard from ARC-AGI, showing Meta in pink, Google in green, and OpenAI in blue:

Then, a peek under the covers of GPT-5, with Mehul Gupta at ‘Data Science in Your Pocket’ decoding the bot’s system prompt.

Next, if you remember recent disturbing news about Meta’s bots and teen safety, they’ve now announced additional safeguards. ‘TechCrunch’ reports:

· The company says it will now train chatbots to no longer engage with teenage users on self-harm, suicide, disordered eating, or potentially inappropriate romantic conversations. Meta says these are interim changes, and the company will release more robust, long-lasting safety updates for minors in the future.

· Beyond the training updates, the company will also limit teen access to certain AI characters that could hold inappropriate conversations. Some of the user-made AI characters that Meta makes available on Instagram and Facebook include sexualized chatbots such as “Step Mom” and “Russian Girl.” Instead, teen users will only have access to AI characters that promote education and creativity, Otway said.

And another look at the looming AI browser wars, with the lead frontier model developers looking to consolidate their hold on your desktop. A few months back, the Bond VC report on ‘Trends — Artificial Intelligence’ did a good job of covering the move from traditional desktop software and OSs to one based on LLMs.

Enrique Dans takes a browser-centric view, and the move to incorporate AI functionality by both Google and OpenAI. He notes:

Chrome is not simply a browser, it is the access point that millions of users open every morning. Whoever controls that controls a decisive part of the digital business. And if it also manages to associate that experience with AI functions that really make a difference, it could change the rules of the game.

Lastly, turning to the hardware side of LLMs, another post by Enrique Dans details DeepSeek’s move to better support Chinese chipsets (i.e., Huawei and Cambricon) at the expense of Nvidia. He explains progress to-date, challenges, and implications.

Of Interest:

· ‘DIGIDAY’ reports on how a creative agency rotated too much to AI, with disastrous consequences

· Tiyasa Mukherjee at ‘Towards AI’ on the reasoning framework evolution, detailing Zero-Shot, Chain of Thought (CoT), Tree of Thoughts (ToT), Graph of Thoughts (GoT), and finally, Buffer of Thoughts (BoT)

· Another good LLM tutorial by Dr. Ashish Bamania at ‘Level Up Coding’

· Rishi Patel, PhD at ‘Generative AI’ with a practical AI agent framework grounded in the real-world data of the 17 UN Sustainable Development Goals (SDGs)