Headline releases in the preceding week were Deepseek-3.2-Exp, Sonnet 4.5 model from Anthropic, GLM-4.6 from Zhipu, Sora 2 video model from OpenAI
Readers who think others in their family, friends and acquaintances who have a curiosity in knowing more about rapidly evolving AI tools/services/use cases would benefit from being subscribers of this weekly newsletter are encouraged to share this publication link to them and invite them to subscribe
For those who use RSS readers to aggregate their reading of news such as the delightful NetNewsWire [MacOS,iOS, iPadOS] or Feedly [web, iOS, Android], this weekly newsletter is also available via the following RSS feed
https://paragraph.com/api/blogs/rss/%40twiata
Below is my personal website which aggregates links to many of my socials as well as content and community that I curate
goolamabbas | Instagram, TikTok | Linktree
Chief Knowledge Officer @ Animoca Brands. Also assists in Growth Marketing
The following messages were posted on the 'All Things AI' Telegram group from Sunday 28th Sep 2025 to Saturday 4th Oct 2025
1 | DeepSeek AI released DeepSeek-V3.2-Exp , as an experimental upgrade to its V3.1-Terminus model, introducing DeepSeek Sparse Attention to reduce computational costs for long text sequences. The model delivers up to three times faster prefill speeds and eight times cheaper decoding at 128,000 tokens while matching benchmarks in knowledge, reasoning, and math tasks. It is now open-source on Hugging Face, integrated into DeepSeek's API with over 50% price cuts, and optimized for frameworks like SGLang. |
2 | Poster: Dima Kuchin Does AI Get Bored?We always give AI something to do. Chat with us, do tasks for us, answer questions, parse text. What happens when we give an AI nothing to do? I didn't know, so I tried. I told it that it had "10 hours" and nothing to do, and to use that time however it wanted.  |
3 | Poster: Vano She has some amazing content and explains very well too. |
4 | Poster: Dima Kuchin Introducing Claude Sonnet 4.5Claude Sonnet 4.5 is the best coding model in the world, strongest model for building complex agents, and best model at using computers.  |
Zhipu Launches GLM-4.6 with Enhanced Coding and Agentic Capabilities and increase in context length from 128K to 200K
Welcome to use my referral link to join their Coding Plan which currently starts at USD 3/month for 120 prompts every 5 hours
You get an instant 10% off on your first order.
🚀 You’ve been invited to join the GLM Coding Plan! Enjoy full support for Claude Code, Cline, and 10+ top coding tools — starting at just $3/month. Subscribe now and grab the limited-time deal! Link：Â
Various threads on X about GLM-4.6 including benchmarks
https://x.com/i/trending/1972980172706951265
1 | Poster: Just | LDA |
2 | Replit acquires OpenInt which powers its Replit Connectors that is being launched today https://www.linkedin.com/feed/update/urn:li:activity:7378880035451023360/ Replit - Introducing ConnectorsBuild Apps & Automations That Work With the Tools You Already Use Today, we're excited to launch Connectors-a new way to build apps and automations that seamlessly integrate with the tools you already rely on every day.  |
3 | One of the clearest explantion of Context Engineering that I've seen. Â Encourage those who are building/creating agents to read this Effective context engineering for AI agentsAnthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.  |
1 | Mira Murati's Thinking Machines Lab launches its first product, Tinker, an API for fine-tuning language models, in private beta with support for Qwen and Llama
Tinker is cool.
If you're a researcher/developer, tinker dramatically simplifies LLM post-training. You retain 90% of algorithmic creative control (usually related to data, loss function, the algorithm) while tinker handles the hard parts that you usually want to touch much less
Introducing Tinker: a flexible API for fine-tuning language models.
Write training loops in Python on your laptop; we'll run them on distributed GPUs. Private beta starts today. We can't wait to see what researchers and developers build with cutting-edge open models!  Announcing TinkerIntroducing Tinker: a flexible API for fine-tuning language models.  |
2 | Aqua which is a voice-input app on Mac/Windows with a SOTA [state-of-the-art] model Avalon is offering a great deal for students who subscribe to its paid plan via their .edu address Students — get 70% off when you sign up with your .edu email |
Perplexity's agentic browser Comet is now generally available to download for everyone (free, Pro and Max users)!
Desktop version: https://perplexity.ai/comet
Mobile will coming in the next few weeks
A prompt engineering guide from Perplexity on how to get the best out of Comet
Here's my prompt engineering guide to getting the best results from Comet:
I welcome feedback not only on this newsletter but across all the various surfaces where I curate content
The cover image of this newsletter via generated via the Mystic 2.5 Flexible model from within the Freepik tool via the following prompt
Floating geometric shapes bloom into vibrant, multicolored flowers mid-air on a sunny spring day, Memphis design with painterly textures, set against a clear blue sky.
Regards, Yusuf