Parameter Update: 2026-03

Things didn't feel that fast this week, but in hindsight there's actually a lot in here!

OpenAI

ChatGPT Go & Ads in ChatGPT

After initially launching in India, OpenAI has rolled out the $8 "ChatGPT Go" plan worldwide. It provides 10x rate limits to GPT-5.2 Instant, but the blog post doesn't explicitly mention 5.2 Thinking. With that in mind, I am struggling a bit to understand who this is for (especially in markets like the US or EU, where the difference between an $8 and $20 subscription is not crazy).

In addition to the new plan, OpenAI has also announced they will start rolling out ads in ChatGPT for Free & Go users, making it seem like an even worse deal.

It's incredible that the enshittification of ChatGPT is already starting this early - especially as they appear to be struggling to maintain market share vs. Gemini as Google keeps giving it away for free?

Either way, here's an example of what these ads might look like.

Elon Lawsuit

New emails have been released in the mud slinging that is the Elon Musk / OpenAI lawsuit. And they are.. not great? for either side.

On the one hand, we have Elon, who was apparently pushing for the for-profit restructuring to finance self-sustaining cities on Mars (?) and wanted to ensure his children would be the ones to control AGI (??), with Altman saying "Elon said he wanted to accumulate $80B for a self-sustaining city on Mars, and that he needed and deserved majority equity. He said that he needed full control since he’d been burned by not having it in the past, and when we discussed succession he surprised us by talking about his children controlling AGI."

On the other hand, we have Brockman, very openly speculating about what it would take to make him a billionaire (and how to get out from under Elon), saying "Financially, what will take me to $1B?" and "his story will correctly be that we weren't honest with him in the end about still wanting to do the for profit just without him" - yikes!

OpenResponses

Ever since OpenAI transitioned to using the Responses API, there has been quite a lot of fragmentation and disagreement about how to handle reasoning traces, tool calls,... - which might finally be about to be resolved this week, with OpenAI (and a load of partners, like Vercel, HuggingFace, and OpenRouter) standardizing around the OpenResponses API. Transitioning the spec from being proprietary OpenAI tech to a community-steered initiative feels like the missing step towards wider adoption, as that was one of the things that really made MCP and Claude Skills work as well as they have.

Claude Cowork

For a while now, my timeline has been filled with people talking about their non-tech relatives using Claude Code for miscalleanous knowledge work tasks. I was very skeptical of these claims, as the form factor of a CLI that keeps prompting for approval to run commands just didn't feel right for people like my mum. This week, it would seem that Anthropic agreed, launching Claude Cowork - effectively a UI wrapper for Claude Code, making the experience much, much more accessible. Interestingly, the whole tool feels vibe coded (because it was - they apparently built it in a week and a half?), but I am not sure that's an issue here?

While I have seen some complaints about smaller UX issues and slightly lower intelligence that native Claude Code, and I am sure a lot of users will still have the usual struggle around use case identification, this feels like a pretty big step!

Google

Personal Intelligence

Memory is one of these ChatGPT features I ended liking a lot more than I expected, as it saves me from having to provide a lot of context every time I open the app. While Gemini has their own version of it, this week they went a step further, allowing the model to dynamically load context from a bunch of Google apps, including Gmail, Photos, YouTube and Search. In their examples, this might lead to the model looking up a car registration from an image saved to Google Photos and then cross-referencing it with an ongoing email exchange.

I would love to try it, but in typical Google fashion the rollout communication was slightly terrible (with people on Twitter sharing various links where you might be able to opt into the test) and it turns out the whole thing is US-only for now. Sad!

Gemini powering Apple Intelligence

It's been very clear that Apple is struggling a lot when it comes to remaining competitive in AI. Fortunately, they have have loads of ressources at their disposal and aren't afraid of simply buying stuff to remain competitive. In an attempt to combat their ongoing struggles with next-gen Siri, then, they have now resorted to licensing Gemini from Google for the next couple years. It's unclear how much money changed hands here, but I assume it must've been a pretty big paycheck.

Thinking Machines Drama

In the final piece of drama this week, Thinking Machines CTO has been terminated due to "unethical conduct". During the company all-hands announcing the departure, two more researchers announced their departure from the company.

The same day, all three of them announced they would return to.. OpenAI! I am not a lawyer, but something here feels illegal? Either way, I will reiterate how excited I am about the netflix documentary that will eventually come out about all of this.