Hello everyone,
As time goes on, AI-generated content seems to be getting more difficult to digest. It’s not just the poor quality; I often find myself wanting to shout, “Please communicate like a regular person.” I suppose that irony isn’t lost on me.
Recently, I’ve been experimenting with two new directives that have significantly enhanced the clarity of outputs, especially when referring to projects I’ve worked on. If I indicate that “I’m not technical,” the system resorts to analogies for explanation—though, honestly, that can often be more frustrating than the initial messiness.
This is an example of how I’ve described a Chrome extension that I developed (referenced in my post from Friday).
Simply add this line to your custom instructions (in settings) or your AGENTS.md file, as they sync with each other:
Always communicate in ASD-STE100 Simplified Technical English.
But wait, what exactly is ‘ASD-STE100 Simplified Technical English’? It’s a standardized writing approach that simplifies technical information, making it easier to read, understand, and translate.
In addition, I included:
Always approach our conversations as if I have ADHD.
The combination of these two instructions has proven to be my sweet spot. The ADHD reference encourages more concise bullet points and headers, creating a dream format for skimmers!
*I wasn’t aware that all the images were transparent, so when enlarged, they looked unappealing. Lesson learned!
On another note, my wife thinks there must be a “someone or a robot” at the other end of her chats.
Ben Tossell
@bentossell
wife’s using chatgpt
“They must find me so annoying.”
Who?
“The computer people. Look how annoying I’m being.”
There is nobody there.
“Yeah, but someone has to be there; even if it’s robots, I feel sorry for them.”
😂😂😂😂
She also keeps using the word “please” each time, wasting tokens.
Ben’s Bites is supported by Cloudera
72% of enterprises believe their current data architecture needs a substantial update to meet AI goals. Discover how this transition is unfolding in Cloudera’s Great AI Re-Architecture report.
OpenAI has updated the availability of its models in the standard version of ChatGPT. Paid users have access to a special version of GPT-5.6-Sol by default, while free users can enjoy unlimited interactions with GPT-5.6-Luna (likely with reduced reasoning capabilities).
Moreover, they have expanded their cybersecurity initiative, Daybreak, and released GPT-5.6-cyber alongside it. Astra, OpenAI’s next mainstream model, is essential for cybersecurity, so we’ll need to wait a while longer for its release.
Bloomberg recently reported on OpenAI’s new device — likely a display-less smart speaker in a doughnut shape. It’s portable, priced over $300, and expected to launch in 2027. (non-paywalled version)
Meta is also returning to open-sourcing models — their new Muse Glimmer model features 30B parameters and is fine-tuned for agentic tasks (including computer use). Meta plans to release the weights for Muse Spark 1.2, their top-performing model to date. Mark Zuckerberg has also authored a 6000-word article about his and Meta’s approach to AI. To summarize, Mark envisions that everyone will have:
-
Personal agents to understand their goals
-
Tools designed for creativity, business, and education
-
Affordable access to these resources
Ben Tossell
@bentossell
I firmly believe that all these will happen simultaneously:
-
More individuals will become builders (capable of creating their projects)
-
A rise in personal software availability
-
Software engineering will transform but will become an incredibly popular career choice
-
SaaS will fall out of favor
-
Years of SaaS and various companies generating massive amounts of revenue will reshape the industry
Grok has launched a new image generation model – Imagine 2.0 which holds the third spot on the image generation leaderboard and second on the image editing leaderboard managed by arena.ai.
Sessions in Claude Code can now communicate with each other.
Here are some additional highlights:
Reliable AI is tough.
Nutrient converts PDFs into JSON or Markdown, providing coordinates and confidence with each value. Grab an API key to try it for free.
-
Raindrop Signals 2.0 employs custom classifiers to identify problems in AI-agent traces with near-frontier accuracy, at far lower costs.
-
Every company needs a Cassandra – a quick read that’s worth the effort.
-
Agent Plugins have become the standard for packaging MCP servers and Skills, now supported by Codex, Cursor, Copilot, and more.
Tareq Ismail@tareqismail
Guillermo Rauch provides insight on reading code. If you’re not doing so, it might be indicative of your skill level, the state of your software, or your business model.
As the landscape evolves, companies need to adapt to the rising costs associated with AI. Every AI action incurs expenses, and existing businesses must consider adapting their pricing models to remain competitive.
In summary, as we advance through this wave of technological innovation, it’s crucial to stay updated and adapt our methods and strategies accordingly.
