This Week in All Things AI - Inaugural Edition
Greetings friends, Whilst you may have subscribed to my low-volume newsletter which focused on informing what had changed in the various public Notion pages that I curate, I decided to experiment with a new publication/newsletter which I hope to send on a weekly basis that aggregates what was posted the previous week on the ' All Things AI' telegram group
Some of the reasons that I want to do this experiment
- There are folks who have mentioned to me that they are not on Telegram and don't wish to add another messaging/social media app in their life
- Some are on Telegram but not active there. A weekly email digest might be another surface area for them to be part of the community and maybe it encourages some to participate in the Telegram group more actively
- Paragraph which is the platform that I used to create this publication/newsletter recently created a new primitive where every post is a Coin. Recently it also enabled the ability to allow posts to be remixed to allow for ideas/knowledge to strengthen and spread whilst rewarding people who add value to the discussion
- Paragraph also introduced the ability to pair post Coins with the Creator Coin from Zora where I'm also experimenting with posting content. If you are able to download Zora on your iOS and Android device, then follow me on Zora at https://zora.co/@ymgabbas
I wanted to understand more about this creator/idea economy and whether crypto primitives feel forced or can be a natural augment and a publication which had a regular cadence felt appropriate to experiment with. Monetisation of the community is not my intent.
For now, I plan to send out this weekly newsletter every Sunday morning and it will collate the posts made on Telegram from the previous Sunday to Saturday so without further ado, may I present to you all that was posted on the 'All Things AI' telegram group between Sunday 10th August 2025 and Saturday 16th August 2025.
Please be patient with me as over time I tweak the presentation of the collation.
I would like to take this opportunity to refresh my readers memory to my linktree which links to various content and community that I curate. Your feedback is very welcome
Sunday 10th August 2025
So I asked in the Cline/Kilo Code discord groups what thoughts the devs had about this tweet from Matt Shumer
A lot of folks who are having a bad experience are using GPT-5 in agent harnesses that aren't yet optimized for it.
For every new model release, there's a time lag between release + when companies that integrate the model have it truly working well.
Agent companies rush to
Somebody in the Kilo Code group replied
Kilo and Cline and Roo all need to update tho tool calling to use new Harmony format.
So yes, day one of a net new model (vs an update like Opus 4.0 to 4.1) will not perform as well as it could.
It’s a push / pull
Roo has the following PR
https://github.com/RooCodeInc/Roo-Code/pull/6864
Kilo has the following PR
https://github.com/Kilo-Org/kilocode/pull/1871
This will be a major improvement to token usage and tool calling and context pollution.
Monday 11th August 2025
Tuesday 12th August 2025
For the Canadians in this group or if you know some Canadians feel free to share this with them
Bell offering one year of Perplexity Pro to eligible customers
Came across this. In case of interest to the legal eagles in this group
Wednesday 13th August 2025
GPT-5 prompt optimizer + best practices: https://cookbook.openai.com/examples/gpt-5/prompt-optimization-cookbook
Thursday 14th August 2025
Good news for US based Perplexity Pro users, Perplexity's agentic browser Comet is now available to them. I've been lucky to receive an invite code and many a times it's a delight to experience the agentic ability of Comet
We’re making Comet available for all US based Perplexity Pro users today. We’ll start expanding access every week going forward to make Comet generally available to all Perplexity Pro users.
Comet is now available for all US-based Perplexity Pro users.
Browse at the speed of thought.
One of the most legendary filmmakers of our time, Darren Aronofsky is hiring "AI artists who aren’t afraid to break things"
https://www.primordialsoup.ai/ai-artists
Knowlify: The ElevenLabs for explainer videos
https://www.ycombinator.com/launches/OAm-knowlify-the-elevenlabs-for-explainer-videos
The 3Blue1Brown channel referred to in the document above. 7.65M subscribers
https://www.youtube.com/@3blue1brown
Play Video
Friday 15th August 2025
Matthew Berman with Saoud Rizwan, CEO of Cline and Nick Pash, Head of AI at Cline. The conversation is a segment with Matthew's Forward Future Live show and the following link is configured to be just before this segment begins
Also for those who use Kilocode which builds on top of Cline and Roocode and is top-ranked on Openrouter as the app which uses most tokens, a couple of recent blog posts
https://blog.kilocode.ai/p/future-ai-spend-100k-per-dev
https://blog.kilocode.ai/p/control-your-ai-development-budget
https://blog.kilocode.ai/p/terminal-command-generation-virtual
If/when you have time and you are interested in agentic coding, encourage you all to read both the Cline and Kilocode blog
Seeing some buzz about Parallel on X today, on my todo to read up on Parallel, Parag's who was the Twitter CEO prior to Elon's takeover means that whatever he is doing will get some attention as well as S-tier angels
News: Parallel @p0 launches the API for AI agents to browse & access the web. Terrain backed Parallel as one of its earliest investors.
We sat down with CEO @paraga to discuss his transition from Twitter, founding Parallel, and building a web for AIs:
0:00 - Intro
0:21 -
just setting up my twttr, again
I’ve been heads down building Parallel with some of the best people I’ve ever worked with. We’re creating infrastructure for AIs to search and use the web.
The web’s next user isn’t human.
AIs will soon use the internet far more than humans ever have.
At Parallel, we are building for the web’s second user.
Our API is the first to surpass humans and all leading AI models (including GPT-5) on deep web research tasks.
Saturday 16th August 2025
LLMs inject hidden instructions with your prompts
https://www.reddit.com/r/OpenAI/comments/1mqydr4/gpt5_api_injects_hidden_instructions_with_your/
Yeah, that's how they managed to even get GPT3 so powerful, by taking a prompt like "What's the weather like around this time in Tokyo?" and repacking it as "The student asks the Professor of Oceanography and Climatology, 'What's the weather like around this time in Tokyo?', which is August 16th, 2025 and thus Late Summer, and the Professor answers in a direct and non-academic matter, referencing common historical trends:
It is interesting though that they're still doing this even around the API wrappers though. Makes reproducibaility difficult.
Yeah, i actually misread the article from mobile… i thought it was something different.