AI industry jargon all over the place. Models aren’t unsafe, they’re misaligned. No betas, research previews. Not engineers or researchers, members of technical staff. What am I missing?

The full suite of what OpenAI's building with ChatGPT is magical. A client has strict requirements for AI use. Precludes many apps. We discussed it in Slack. Dropped my Dot in the thread and asked for a ChatGPT Site that uses only approved models. Done in under an hour. There were a few hiccups, and typical AI artifacts. First, it got confused about incorporating an OpenAI API key, telling me it could only be done server side, whereas I can and did add one manually in site settings. Second, I asked it to incorporate WebMCP so agents could use it easily, and it wasn't able to properly test this was working. Third, there are lots of little unnecessary text elements and callout boxes all over, all that self-referential stuff that AI loves to add. But overall, pretty awesome. We hit a barrier, discussed solutions, sent my Dot to build something custom to solve the problem, and it built and launched it in less than an hour.

Had Claude polish the design and develop a design language for infographics associated with news items, which Muse then implemented flawlessly. Looking good. Now I just wish I could get the "Discuss" button working (and ideally get a cleaner, more native-looking sidebar icon).

Annoying bug in Muse right now: when it uses buttons, the text it sent along with those buttons doesn't come through, so you only see the buttons and have no idea what they refer to. Filed in Muse but keeps popping up so hopefully someone can jump on this.

Muse feed is good but coverage is light, about one item every two hours. I had Muse build a custom news artifact to augment it. Scans Hacker News, Techmeme, and high-signal X accounts every hour, filters against inclusion criteria, alerts me to anything urgent based on alert criteria. So far, it works really well. It shortlists posts every hour and reads the ones that pass its initial filter, then writes up summaries. It stores everything in whatever database artifacts use (🤷‍♂️) so they're searchable. I'd like to improve the visuals that accompany posts, including by incorporating images associated with the items, and polishing up the infographics. I'd also like the ability to click "Discuss" on an item and discuss it in the chat with my Muse, but it's not clear this is possible. Muse tried it initially, based I guess on how it thought it should work, and it didn't.

Does anyone know if it's possible in Muse to incorporate the "Discuss" functionality into an artifact? In the feed, you can click "Discuss" and add that context to the chat to discuss it. I want to do the same thing in an artifact. My Muse couldn't figure it out.

Gated access to proprietary cyber defense models feels increasingly untenable as open weight models match and exceed what proprietary models could do just months ago.

While I still prefer Muse for most things, Grok Bot is (perhaps obviously) exceptional at everything X-related. It alerts me to breaking AI news on X judiciously, minimizing noise, and monitors posts for replies upon request (like the Meta feed issue I just flagged).

I've built the same Spanish learning app as a Muse artifact and a ChatGPT custom plugin built by my Dot and backed by ChatGPT Sites with an MCP server. (Left is Muse, right is ChatGPT.) Both work, but Muse (initially with Spark, later with Wonder) is better for speed of implementation, speed of iteration, and speed of the app itself. Muse built fast and flawlessly for this app (no major bugs), finds good opportunities to streamline the interface when I ask, and acts on my feedback quickly and thoroughly. This speed of implementation and interactivity matter: I find I learn better from the Muse version because the app itself is snappier, and my feedback gets implemented faster. The ChatGPT build still got to a good place. I had given my Dot requirements and CSV files from the original app I hacked in Muse, and it rebuilt it pretty well. I do feel like I was initially asking it to remove unnecessary text and distractions to streamline the UI more than with Muse. But the biggest issue for me is speed. It takes a lot longer to make changes, and the app itself is way less responsive. Now, if I was going to scale this into something I made available to a larger number of people beyond me, I think ChatGPT Sites might be the better approach. It has authentication built in, for example, and is almost certainly more scalable. But for a personal app, just for me, I prefer the Muse artifact. Some other notes and thoughts: - For both apps, I asked the models to use the native styles for my selected themes. Hence why they both have a similar color scheme. - Yes, I could build artifacts in ChatGPT (or, at least, this used to be possible; it seems like plugins are the preferred approach now). But you can't add them to the sidebar and open them in full screen mode like a native app the way you can in Muse, as far as I know. You can only do this in ChatGPT with custom plugins using extensions.

Using Muse today, it made an inline chart. I’d never seen that so asked about it. Maybe new? It says it has a bunch of widgets, but isn’t sure if they’re new. These include custom visuals and tools that it says maintain state. This makes it more competitive with ChatGPT’s chat mode. Note to readers from Meta: this is fantastic, and I love that the colors inherit from my chosen theme. Nice touch. Three issues I noticed though: (1) If your theme uses a light color, some things rendered in that color like chart lines and headings can be hard to read. You should consider complementary colors for widgets that reflect the theme but will always be legible when used. Or, better: every theme can have widget colors prespecified and then these can be checked for legibility (I imagine eventually people will be able to create custom themes, so having values for widgets may be worth considering). (2) Initially the chart widget lacked margins. Text ran right up to the edges. My Muse compensated by adding padding inside the widget, but the default should be fixed. (3) If you click on a widget and try to share it, it shares a blank box. The widget content doesn’t show up in the preview and I assume also when you share it.

We need the ability to completely customize the left sidebar on the ChatGPT iPhone app, especially now that we can create and install plugins that can be added to the sidebar. I want my News app link to be prominent, not buried behind an extra click.

Something cool and useful that Muse does and other personal agents like Dots don’t seem to is background sync local apps like Apple Calendar and then use that synced data to do things like proactively prepare you for meetings. Just now, my Muse proactively drafted a document to prepare me with talking points for a work meeting later this week. But it doesn’t have cloud access to my work calendar, only to my personal Google Calendar. How did it know about that meeting? Well, it has access to my Apple Calendar on my phone and MacBook, and my Apple Calendar includes my personal, family, and work calendars. Turns out those get synced so Muse can use the synced data for background tasks.

Another activity that’s better in a Chat thread than with Dots in ChatGPT is visual brainstorming with image creation, and I think brainstorming in general. Chat responds faster, writes better, can do more visual output, and provides more purpose-built controls for image editing that aren’t available from Dots. In general, I keep finding that Dots are great when you know exactly what you want and can delegate it and have your Dot babysit the task and ensure it goes well. This includes scheduled actions. But when you want to learn, discuss, brainstorm, or generate images, Chat threads are best. And I still find Work threads are just not something I use much anymore, with the exception of projects where I need consistent context or files and often use a Work monothread.

A both good and bad Grok Bot experience this morning. The good was it proactively found a developing story on X about the new leaked OpenAI image model endpoints, which my other AI news feeds didn’t. The bad was that yesterday it created a bot specifically to monitor X, and despite what it says, we gave that bot an hourly schedule, but it perhaps wasn’t ever set. I’ve found where scheduled tasks live now and am going to inspect those to make sure bots are set up to do what they’re supposed to.

Current personal agent ranking: - Muse is the most user-friendly, polished, personable, proactive, and enjoyable to use. It also offers unique features like goals and podcast generation. But I don’t fully trust it with my most complex tasks and research requests, and I really want voice mode. - Dots are the most capable and offer all the benefits of ChatGPT, including the awesome new chat model with intelligent UI, Sites, and custom plugins with extensions. They also integrate into Slack beautifully. But they’re slow, rough around the edges, opaque as to what they’re doing, not as proactive as Muse, and quite dry in terms of personality. - Grok Bot is fast and you can create many bots. Also, you can give bots email addresses, although sending from those addresses requires per-email approval, which limits their utility. But privacy options are weird (no training on your data also means bots can’t send feedback to SpaceXSI on your behalf, for example) and bots hallucinate features that don’t exist. It’s early for me with Grok Bot, though, so we’ll see how my feelings change over time.

Surprised OpenAI hasn’t bought Reddit. Would expand ad reach, bring real-time news and product reviews they could cut others off from, and give them parity with SpaceX, Meta, and Google with a social network and distribution platform.