Happy to announce @storycraftgg has raised a $3M seed round from @khoslaventures! This adds to our earlier $2M pre-seed from @LondonVP @SignalFire @relayventures and @spacecadet. We’re hiring, DM me for game developer, UX, creative director, & growth/ua roles! Full link 👇

I ain’t buying “personal assistant” as THE consumer ai use case. Yes for VCs and the optimization pilled, but people like designing their daily outfits, and having an “in” at a restaurant, and leisurely shopping as entertainment, and dreaming about their trip while crafting the perfect vacation. These things aren’t preventing people from achieving their dreams, they literally ARE their dreams

Players losing sleep playing @storycraftgg. Sorry / Not Sorry 😂

It’s wild how many model and tool providers exist and how few successful app layer ai companies there are. So many images and videos and so so much code being written but virtually no new experiences.

The medium is the message, and the medium of the web is information not content. Your one shotted astra game is information not entertainment. Hence why the only place AI web games are blowing up is X. Not saying implications aren’t big, just saying folks aren’t thinking about them the right way.

Players of @storycraftgg start by building a game world, then they turn it into an illustrated story, and now they can turn them into movies. AI doesn’t replace creators it’s going to create millions more of them.

One-shotted “games” are to foundation models what Jeopardy was to Watson: marketing, not proof of capability. Legible, but not real. No one is playing them except their makers. Why? Because they’re not actually games.

Offering some feedback to the @ChatGPTapp team in the hopes concrete feedback helps them fix recent issues. The primary value ChatGPT gives me over claude code is the persistent list of topics in the sidebar which represent ongoing long form areas of investigation. Being able to return to them with new context and learnings and continue iterating is possible in ways that the CLI of claude doesn't let me do as easily. Sure I could use the app, but the lack of context between CLI and app prevents me from doing this, and I'm not paying twice (yet). However, here's my recent experience. I've stopped using the desktop app after the work/codex changes because my sidebar context is gone. So I've reverted to the mobile app for this alone. But, I'm having a bunch of new problems. I opened a voice chat and went for an hour walk. I had previously uploaded multiple different documents to the chat that had important context. While walking ChatGPT informed me it could not retrieve these previously uploaded documents. But I had to determine this, it pretended as if it had full context, essentially misleading me until I made it explain what it was using for the analysis I asked it to do. So, it lost trust, then I had to figure out how to upload these docs again while walking around, greatly lowering the value of voice chat as a thought partner because I had to jump to text mode, upload the docs, then start voice chat again. Now the really bad part. It churned for 40m working on my problem, I was excited that it was taking so long as it had not really done this before and so I felt I was going to get a really high quality result to make meaningful progress on my problem. Unfortunately after it failed to return on time I opened the app and saw that I had run out of voice/model tokens on my Pro plan for the day. No warning, no voice explanation this had happened, and no interim results. The next day the model had no way to reconstitute the results from the prior day so I had to start over again from scratch. This time as I uploaded the docs for the third time I would get a popup asking me to switch to the higher quality voice model. This has nothing to do with files but hey why not? Except the new model has no memory of the conversation with the last model. But I don't think of a model as my chat, I think of the chat thread as my chat. Swapping a model should not lose context and force me to start again. So now my only option is to work in chat mode not voice mode because that's the only way I can retain prior context and upload files. Now, in the end I'm getting a consistent reproducible orange error that says "Error in message stream. Retry" that never works after multiple retries. So now I'm stuck. Not for lack of trying or patience. At this point the switching cost to simply moving to the Claude app and rebuilding all the context I have with ChatGPT is worth it because my beloved and much used ChatGPT threads are 1) no longer accessible in the desktop app, 2) bugs are preventing me from progressing in this long form chat, and 3) the voice mode bugs limit a primary use case which is to use ChatGPT as a walk and talk though partner. For now I'm off to see how the Claude app works. Not what you'd want to hear I'm sure, but as a founder and builder I wish folks who love my product would take the time to provide feedback like this instead of just complaining so I hope this helps somebody over there understand the issues that have recently emerged!

The X comments are full of praise but the YouTube comments near universal vitriol. Nobody seems to be reviewing it through anything but their own idealogical lens. It’s a cool concept, shorts are notoriously hard, the production design, vfx, and creature effects are pretty great, but it also feels like Blomkamp fan-fiction, with the acting being forced, the grittier moments looking too clean, and the whole thing just lacking his distinctive style. My guess is they’ll get there if the frontier labs listen to real artists and make the tools work in ways that help them realize their vision. I propose “The Blomkamp Test” in which a real auteur filmmaker’s AI works are indistinguishable from their prior non-AI works. We ain’t there yet!

So I reverted to the web version of ChatGPT, saved it to my dock which kinda feels like I have the old app back, but have now discovered that long chats grind to a laggy impossible to use mess in the web interface. For the love of god @sama please give us a chat | work | code toggle in the new app so we can have the original clean OG ChatGPT interface with the history sidebar! 🙏

Perfect post 🤌

What happened with the new ChatGPT app? Chat history gone, nonsensically complex interface. One of the simplest and most successful interfaces of all time replaced with Jira-like overlapping windows, tasks, projects, popup approvals - my first chat was about how to re-install the previous version. Brutal.

A newly discovered disconcerting trend in hiring: AI idiots. They can pass interviews and homework because they have access to AGI, but on the job they are unable to handle realtime complexity, discussion, and disagreement without the crutch of the AI who is doing their actual work. They instead become triggered when faced with illegible messy real world problems and quickly flame out. Modern screening needs to account for this. We used to have to be on the lookout for the smooth talker but now we have to be on the lookout for the smooth prompter!

Your every five years or so reminder that “ai is the new ui” and “voice is the most natural interface” sound right but aren’t in a bunch of subtle ways (no one wants text or audio only anymore than they want to have ears without eyes, and what we think of as voice is actually a lot of body language, micro-expressions, and context). These aren’t new ideas or even phrases!

I worry about how many young pre-pmf startups are spending time making their company “legible” and building internal ai loops that have nothing to do with finding pmf. There are no shortcuts to talking to customers, and watching them use your product. Automating customer feedback intake and auto generating PRs sounds insane to me as a pmf hunting strategy. At some scale this all makes sense, but pre-pmf it just seems like AI enabled “playing house”

A beautiful article! We prompt because we feel…

The thing that isn’t widely understood about ai creativity is “prompt to x” is DOA because one choice isn’t enough. As George Saunders says on @RickRubin’s @tetranow podcast: “Being a great artist has something to do with creating the maximum number of choice points. Because, if you’re choosing one time that’s one dollop of you, but if you’re choosing a thousand times over the course of the thing a lot of you is getting in there” AI platforms that let creators make a lot of small choices and to revise their past choices are the ones that ensure the person on the other side can feel the person who made it. Prompt to x is slop. But thousands of prompts over days of work can be art, if the definition is that the artist is successfully conveying some essential part of themself through the work.

Slopocalypse Now

One of our @storycraftgg players posted a set of cozy holiday interiors she made today, and I’m missing Christmas already! 🎄

All these folks saying “look I made a multiplayer shooter in 3m with AI” are forgetting to ask themselves why anyone would want to play it when Fortnite, CoD, Valorant, etc exist and that’s where people’s friends are already. True for most “I made game X in Y minutes using AI”. Why would anyone care how long it took, and why would shorter be better? Gamers like stuff that has love, skill and humanity put into it!

There’s a startup angle here - stay with me: The two murderers in this article viewed themselves as entrepreneurs and influencers. A study cited says 73% of Gen Z’s want to start a business. Some ChatGPT work indicates that in the US 50% of Gen Z’s, 41% of Millenials, 31% of Gen X, and 19% of Boomers classify themselves as entrepreneurs. This doesn’t account for retirement, but it also seems like the desire to start a company is going dramatically up. Makes sense given good entry level jobs are harder to come by. How do we then square this with the rise of brainrot startups like friend and cluely that promote style over substance aka marketing over product or VCs that promote momentum over PMF? It feels like the desire to start a company is intersecting with rage bait native founders, coupled with “momentum matters more than product” beliefs in VC and the ability to rapidly build a prototype using AI are all coming together in an uncomfortable way. Get rich quick entitled immorality seems like a point of commonality between both these murderers and brainrot founders and those that fund them. One is clearly worse than the other, but the belief systems seem adjacent. If believing there’s no substitute for empathy, creativity, and hard work is the new shaking your fist at the kids on your lawn then I guess I’m ok with that 👴🏻