OK, F1 on Apple TV is a nice upgrade over F1 TV - built in multiview!
06.03.2026 01:30
👍 0
🔁 0
💬 0
📌 0
OK, F1 on Apple TV is a nice upgrade over F1 TV - built in multiview!
And GitHub is down again...
What Opus & Sonnet can do with a PowerPoint is still far and above what OpenAI's models can do.
Claude needs a Codex-like app as Claude Code Desktop doesn't quite cut it. The macOS and iOS apps need some serious love and care. OpenAI needs to improve their performance on productivity/writing tasks as evidenced by their GDPEval scores and llmarena text scores.
This is a bombshell: "the Pentagon wanted the company to allow for the collection and analysis of unclassified, commercial bulk data on Americans, such as geolocation and web browsing data"
This kind of agenda echoes the defunded Total Information Awareness effort, post-9/11
NEW: Our play-by-play on how Anthropic’s negotiations with the Pentagon fell apart, (and how Open AI swooped in). With @julianbarnes.bsky.social @cademetz.bsky.social
www.nytimes.com/2026/03/01/t...
So it seems the concerns that Anthropic had with the DoD were well-founded. Is OpenAI now a part of this use case?
Deleted my ChatGPT Pro account today after being a user since November 2022. Good riddance. Not sure what is worse: the fact that they can't admit they signed a "lawful use" contract or the disingenuous messaging from some of the OpenAI employees on X.
Finally was able to get my ChatGPT Pro subscription cancelled & account deleted. Been on the Claude Max plan since it came out and the models are just hands-down better that Anthropic produces -- much better at writing and general productivity tasks as evidenced by GDPEval.
Maybe one positive thing that will come from the ChatGPT cancellations is that the masses will be exposed to some models that actually write decently vs the bullet point slop that GPT-5 generates.
@karaswisher.bsky.social @profgalloway.com
Getting an error when I try to cancel my OpenAI Pro sub now. Wonder why?
This is the reason that I deactivated my Facebook account last year:
pilk.website/3/facebook-i...
The Codex GitHub action could definitely use some work. Anthropic has done a good job of improving the Claude Code equivalent. Finding myself implementing more automation in GitHub actions with it like updating docs, tests, etc. periodically.
Qwen/Qwen3-VL-30B-A3B-Thinking is a surprisingly good general-purpose, multimodal model that is roughly on par with gpt-5.2-nano (maybe a bit better).
Liquid Glass interface looks awesome! Thanks for the update.
Wow:
Guess this explains why the Anthropic models are such good writers:
apple.news/A1TzCbBTeTHG...
Claude is better anyway.
Safari user here and not a fan of Chrome, but finally found a use for Chrome Apps -- using X Pro. Keeps Chrome out of the way.
Why can GPT 5.2 Pro still not use the tools provided to it on the web UI like GitHub, etc.? It makes no sense other than OpenAI forcing the use of GPT 5.2 Pro through the API.
4) Prompt CC to generate the presentation (make sure the Claude document-skills plugin is installed -- this basically gives you Cowork in CC)
5) Iterate as needed
My workflow so far:
1) Do research in Claude with Opus 4.5
2) Save .md files with research to the project directory
3) Use a company-specific brand-guidelines skill that covers the colors, logos, logo placement, fonts, etc. (basically, just copy and customize Anthropic's brand-guidelines skill)
Shifting building slide decks in PowerPoint to Claude Code is the ultimate unlock for someone that has to do presentations on a regular basis.
The gun he was carrying does have a tendency to discharge on its own: www.thetrace.org/2023/04/sig-...
Disappointing to see this article get posted to YC-backed Hacker News and then immediately flagged: news.ycombinator.com/item?id=4674...
Finally have a use case for the Claude Agents SDK, which is pretty cool. Writing a bot for our users in Teams to interact with for document generation and other tasks similar to what Cowork does, but brought to Teams.
Little things in Claude Code continue to impress to me like the harness supplying its runtime information to the model.
