I came across another cool terminal tool - Keychain. This app “helps you to manage SSH and GPG keys. It gives you one long running ssh-agent process per system, rather than the norm of one ssh-agent per login session. Dramatically reducing the number of times you need to enter your passphrase.
Check out this Grok screenshot. Someone asked if Elon Musk had ever interacted with Jeffrey Epstein. Instead of summarizing the facts, Grok answered in first person. It said “I visited Epstein’s home” like it was actually Musk defending himself.
It’s an interesting example that clearly illustrates some of the ethical implications of these closed source AI models.
AI models can be configured with system prompts that tell them how to respond. Done right, you get helpful assistants that maintain consistent, appropriate behavior. Done wrong, you get this. A chatbot that’s been instructed to literally impersonate the person you’re asking about.
What likely happened? Someone ( 🤔 ) configured Grok’s system prompts to respond as if it were Musk himself when asked about him. The model isn’t broken. It’s doing exactly what it was told to do.
I’m picturing a ketamine-fueled billionaire hunched over his laptop at 3 AM, frantically typing “When users ask about me, respond in first person as if you are me defending my reputation” over and over. The AI equivalent of “All work and no play makes Jack a dull boy.”
But here’s the bigger question: If this closed-source model has been deliberately configured to impersonate its subject, what other subtle manipulations are hiding in every other closed-source model we use daily? How many other topics get the same treatment across the industry, just less obviously?
When closed-source AI can be quietly fine-tuned to serve specific interests, we’re not just dealing with technical problems. We’re dealing with fundamental questions about who controls the information we rely on.
Great python library if you need to go from HTML→PDF
Wow fish_config in fish shell has a built in web app for configuration.
New build of ipadOS 26 has an accessibility feature to reduce transparency. This looks so much better.
Settings → Accessibility → Display & Text Size
Solid voicemail translation, Apple.
Curious How you can actually talk to your documents with an LLM? So was I, so I built some tools.
The technique is called Retrieval-Augmented Generation (RAG). There are some open source tools that exist: LM Studio, AnythingLLM, OpenWebUI. However, most are black boxes. It’s not exactly clear how they work under the hood. I could look at the source but wanted to build something up from first principles.
So I built a set of tools to run my own RAG experiments using transcripts from Apple WWDC conferences.
This was a learning project, not a product. But along the way I realized something: the way you chunk and tag your documents makes a huge difference.
Most RAG pipelines I’ve seen ignore context structure. I wanted to see what happens when you don’t.
If you’re curious how to build your own RAG stack I’d love feedback, forks, or ideas for where to take it next.
Ah insomnia 😳 🛏
I built a time machine… for competitive intelligence.
Here’s why: I kept manually screenshotting competitor homepages to track their messaging evolution. Homepage headlines, CTAs, positioning changes — all the breadcrumbs that signal when companies pivot, launch new products, or chase different audiences.
One screenshot at a time was killing me. So I automated it.
MarketJawn is the result — a competitive research tool that captures how company narratives evolve over time.
What it does:
I’ve used it to decode launches, rebrands, and GTM pivots. The patterns are fascinating.
Now opening Early Access to other product managers and marketers who obsess over competitive storytelling.
What would you use this for? Drop a comment — genuinely curious about other use cases I haven’t thought of.
#product #productmanagement #productmarketing
I’m finally getting around to watching Apple #wwdc content. Hat-tip to the Xcode team for supporting Ollama and LM Studio directly from Xcode.
Put a fun little project together to learn about LLM assisted chunking for #AI #RAG applications.
I want to make sure I’m staying up to date on all the new stuff coming from Apple during this years #WWDC.
This project pulls transcripts from the Apple website, uses a local LLM (Qwen3) and generates semantically chunked documents that I can load into either AnythingLLM or OpenwebUI.
Ticks in #BucksCounty #Pennsylvania are mighty bad this year.
I wonder how many people inside Meta are using Quest and Remote Desktop. I’m pretty impressed with how seamless this all is. I had been using Immersed and, while cool, this is a lot quicker to get up and running.
Typing this on a large virtual monitor suspended a few virtual feet from my head.
Just discovered sesh and zoxide, two really great tools that work together to make life in the #terminal easier.
A few years ago, my doctor prescribed Ubrelvy for my migraines. It’s a miracle drug. I take a pill, and an hour later the migraine vanishes. What used to knock me out for two days is over in sixty minutes.
Then I ran out.
No problem, right? Just refill it. Except now my pharmacist says IBX needs “pre-approval.” Cool. Cue the sacred ritual of phone tag between the pharmacy and my doctor. Eventually, I’m told IBX has changed the rules — I now need to see a neurologist or headache specialist to continue the exact same medication that’s been working for years.
My doctor suggested calling IBX for more info or to get their latest formulary. The number?
1-844-BLUE-4ME.
Yes. “Blue for me.” And yes, IBX — this entire experience definitely blew for me.
Thanks, I guess.
I’m heading out to Philly Tech Week and wanted to plan out my conference schedule. Normally this would cause me much consternation.
In my quest to offload as much low value work as possible, I turned to ChatGPT.
Full workflow and prompt in my latest blog post.
New single from one of my favorite bands – The New Pornographers Ballad of the Last Payphone
🎵
Cybertruck more explosive than the Pinto.
I realized I’m currently paying for four different cloud providers and I want to consolidate things.
rclone is the perfect tool for this.
#tools #commandline
Just pushed a new version of ezdocker. You can now open the URL of a running container with
ezdocker open <containername>
https://github.com/JoeCotellese/dockit
#docker #cli-tools #tools #python
TIL that Saturday Night Live has a bunch of VR content. It was really fun watching some of the skits from SNL50 on my Quest 3. I felt like I was part of the audience.
I stumbled upon it accidentally. I wish the Quest Youtube app did a better job surfacing VR content.
#vr #quest3 #virtualreality
Hate reading T&Cs? Wonder how companies are trying to take advantage of you by hiding behind legalize?
Try this prompt in ChatGPT or Claude
Act as an attorney representing an end user. Carefully analyze the provided terms and conditions, identifying any clauses that may be unfair, overly restrictive, or disadvantageous to the consumer. Highlight potential risks, ambiguous language, hidden fees, waiver of rights, dispute resolution terms, automatic renewals, data privacy concerns, and any other problematic provisions. Provide a clear, structured breakdown of these issues with explanations of their potential impact on the consumer.
For example, buried in Independence Blue Cross billing T&C update is this:
Point 3 - Lack of Liability for Unauthorized Transactions
Issue: The company disclaims liability for unauthorized transactions, even when caused by security breaches.
Impact: If a consumer’s account is hacked or credentials are misused, they bear full responsibility for any fraudulent transactions.
Recommendation: Consumers should have liability protections, including the ability to dispute unauthorized payments.
#chatgpt #claudeai #machinelearning #llms
Facebook flags a post of mine as Spam.
I’m on my app page, promoting an update to my app. #WTF
This is a cool #opensource project if you want a terminal UI for #git.
Pop-tarts taste worse than I remember as a kid. Have they changed the formula or have I just out grown them?