quinncom

Don’t forget, the middle seat gets to use both armrests. It evens out.

> *Addressing abusive behavior toward our models* > We’ve added a prohibition on sustained and needless abusive or cruel behavior toward our models. The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose. It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.

They’re both excellent, but don’t you find the free quota is exhausted after just a few research queries? I haven’t tried them in awhile, so maybe they’re better now.

You can add Kimi to this list: it has one of the best deep research agents, but just one search uses ~75% of a weekly quota.

Well, there’s this: https://www.openui.com/blog/oui-1

[HEX](https://github.com/anomalyco/hex) is a nice open source app, although it’s in the middle of a rewrite into rust and cross-platform support is still in progress.

The onboarding worked well, but it ended with me in what looked like an AI chat named “MYCOMPANY engineering co…” with two participants. [1] I wasn’t quite sure how to proceed. I guess the engineering comes from the selection I made for type of work, but I wasn’t sure if this was just a chat UI or if I was already communicating with a coding agent. It might be helpful to auto-submit a “What can you do for me?” message to get the conversation started.

Another confusing thing on the screen was that in one of the chat controls there was a menu named “AI Prompt” with two options:

- Al Prompt (Product & engineering will reply) - Team message (Product & engineering will not reply)

What is “Al Prompt”, “Team message”, and “Product & engineering”?

Under workspace settings, I can see that my account has already burned 270,438 tokens, even though I have sent zero messages. [2] Is that from just the system prompt?

[1] https://s.strco.de/f/img-9592.png [2] https://s.strco.de/f/img-9594.png

I love that it has a feature to prove images came from the real world:

> Users can now prove the authenticity of a photo taken on iPhone 18 Pro models with Apple Reference Image, powered by the new sensor in the Main camera that can sign every pixel it sees. When a photo is taken in the new Reference mode, the camera captures signed sensor data that Private Cloud Compute develops into an unalterable reference image.5 Reference images can be viewed in the Photos app alongside the main image, like a digital negative, to visually compare the two assets and determine if any edits were made. APIs are available in iOS, iPadOS, and macOS 27 for third-party apps to enable viewing of these reference images.

I agree on both: indeed a nice article, and I'd like to see a chart with speed as x axis.

As someone who only needs AI for a couple of tasks per day, I don't really care how much it costs, especially when subscriptions are subsidized. I want to filter by speed (eg, max task time < 1 min) and then choose the intelligence I need for the task. This will surface models like Gemma4 31B (xhigh) running on Cerebras and GPT Sol (med) fast mode. Using these models feels great and are affordable for infrequent tasks.

Looks fine in my Firefox. The text size is smaller than in Safari, but it’s fine.

Alt link because archive.org is down now: https://archive.ph/V8H5s

I’m blocked on the phone number registration step:

> Enter a valid UK mobile number (e.g. 07700 900123)

(I have only a Google Voice number from the US. I spend a lot of time in London, so feel like I should qualify for an account. )

Initial benchmarks on my M1 Pro, about 10x faster at describing screenshots than qwen3.5-4b. If the quality is comparably, as the model sheet implies, that’s incredible. But it seems this isn’t a reasoning model?

You can also submit to the EasyPrivacy list or one of the AdGuard lists.

I only get ~4 tok/sec on a M1 Pro with MTP.

AI;DR

> That distinction matters when you're debugging an agent, because "the answers got worse after the model swap" is not one failure, it's five

Freedium now supports FT: https://freedium-mirror.cfd/https://www.ft.com/content/1b0a8...

You can also add comments directly in the diff view and they submit in the next message you send.

Caching should be possible as long as all the models use the same shared cache. The models don't even need to be running on the same server if the shared cache is distributed.

I have a feeling people reading this are thinking that a model router would be used to route between different providers. And in that case, a shared cache would be impossible, although some caching would still be effective. I think, ideally, a router like this is in front of a set of models hosted in one place.

Raising the is–ought fallacy here i s a category mistake that weakens the author’s argument.

“Software is about people” argues:

> I used to think software was fundamentally about code. Experience taught me that I was mistaken. Software is fundamentally an artifact made by people, for people’s purposes, and its success depends on them.

“Software ought to be about people” says:

> Software isn’t necessarily about people, but I believe it should be.

You may argue the latter, but I think the OP believes software is already shaped by humans .

We’ve been in the singularity since the 1950s; exponential grown can be flat for a long time.

Or, as the Wikipedia Simplified Technical English Wikipedia mentions, use Plain English or Plain Language as non-aerospace alternatives:

> The US government's Plain English lacks the strict vocabulary restrictions of the aerospace standard, but represents an attempt at a more general writing standard.

https://en.wikipedia.org/wiki/Plain_English

> Modern corporations and academic organizations hire technical writers already trained to follow a simplified form of English that the layman reader can easily understand, known as plain language (PL).

https://en.wikipedia.org/wiki/Plain_language

And so we've looped back around OP’s “plain English” phrasing.

ChatGPT Group Chats were retired July 9.

https://help.openai.com/en/articles/12703475-retiring-group-...

Excellent, problem solved.

Update after using the new ChatGPT.app for a bit: Lots of old ChatGPT functionality missing:

  - No voice mode
  - No Deep Research
  - No explicit Search toggle 
  - No Record mode (meeting summarizer) 
  - No Canvas (good riddance) 
  - No custom GPTs 
  - Old chat threads are hidden (have to open a new chat, then click “See all” under “Recent chats” – or use the “Search Tasks…” function)

Does the new ChatGPT app still separate ChatGPT and Codex usage quotas, or does all usage now count against the Codex limit?

> The updated ChatGPT desktop app (opens in a new window) is available globally today for Mac and Windows, with Chat, Work, and Codex available to users on every plan, including Free. > If you already use the Codex app, you can update it as usual and it will become the new ChatGPT desktop app. Developers can make Codex the default view when they open the desktop app and choose the Codex logo as the app icon. Desktop Codex projects remain accessible on the go through the ChatGPT mobile app. The existing version of the ChatGPT desktop app will be renamed ChatGPT Classic.

---

Edit: I just got the update – the new ChatGPT.app is 1.46 GB

Something not mentioned in the changelog that I just noticed: Kagi Assistant now offers Gemma 4 31B hosted at Cerebras. It’s insanely fast and seems to follow instructions quite well as a research assistant.

I just got it working with Pi, it works great.

I use Pi inside nono.sh sandbox, which requires setting:

export PLANNOTATOR_REMOTE=1

Then adding to the nono profile:

{ "filesystem": { "allow": [ "~/.plannotator" ], }, “network": { “open_port": [0], "listen_port": [19432] }, "open_urls": { "allow_localhost": true } }

If I want to onboard a client to Cloudflare, I have to ask them to create an account and then invite me, which is a lot of friction for non-technical people.

A “create account” button accessible to me would be so much better. Then, I create the account and invite the client to join as owner.