AI Tweets You Missed This Week - 28th June

Everything I bookmarked that is worth sharing about AI in the past week.

Nothing on the creative front this week, apart from a really bad Dave Chappel deepfake. But I'm also finding some weird innovations that defy previous categories, so let's start there:

Wrapping an LLM into a font is pretty mind blowing. First, what the hell else are people rolling into fonts, that sounds like a security concern. Second, does this mean apps don't have to update with AI features if you just need to load an LLM font into the app to get the same features? This is the underrated innovation of the month IMO:

Wifi signals reading room spaces is such an incredible hack on existing infrastructure, and such a mindf**k on the modalities AI can work with, that I think a big part of the impending luddite movement is rejecting technology we've otherwise been comfortable with. Will people let this stuff into our homes when this is mainstream?

It gets worse:

Unrelated to AI specifically but another example of a modality you never imagined giving intelligence a human mind could never interpret. (Years ago there was some study about working out a PIN on a phone or similar based on the ambient light changes from your hand being read by the camera. Cyber crime in the future is going to be NUTS)


I need to dig more on Websim's injection feature, because I can't get my head around how people are getting this level of interactivity out of pure URL prompts:

More audio experiments. Let's build an entire studio workflow in websim!

Hilary keeps banging out smart prototypes with Websim:

Hliary's demo is similar to another banger from Websim this week. This is an amazing tool to explore concepts, and you can share the result. I was only three clicks from my favorite topics on mass behaviour and mimetic engineering:

Love the result from this game built in websim, mixing up Tetric and a tamagotchi (which you might recall from last week's newsletter):


Yohei brushing up BabyAGI, one of the founding agent platforms:


Fed up of debates? Have two LLMs fight out a topic for you 😄

Love this idea on building scoring and evaluation tools for decision making. All to often I build ranking tools in a Google Sheet or similar to help rank ideas etc - but you can get bogged down in building a "perfect" scoring system. Tools like Claude's Artifacts get all that out the way so you can get on with it:

Truth and fact is probably the biggest challenge and goal facing big AI companies. This research claims to make big improvements:

This is apparently state of the art lip sync. Still loops a bit ropey to me, but you can imagine the creative potential:

MultiOn (not the Multi bought by OpenAI) looks like a killer scraping tool using just natural language. I've tried "easy" AI scraping like BrowserBear and still battle with it, so I'm looking forward to trying this one:

Automating intelligence is my holy grail for AI tools and generally the focus of my private R&D. This is a stellar example of a system that, in its output, throws up the info you need to wake up to to be ready for the day's meetings:


Agents as microservices?


One big overlooked use of AI (outside of the big companies who do this routinely) is models for pure evaluation and standard checks for faster feedback loops. Personally I've been wanting to make something that vets our company content for whatever standards I intuitively have when approving content personally. Here's a great example of OpenAI using a model to supervise its own code:

Has anyone figured out what this is yet? The logical reason for S2S is probably 'speech to speech':

While I'm generally averse to SaaS platforms when everything is essentially a wrapper that could be done privately or open source, there's sometimes a proposition like this that cuts through and feels worth a look as a business owner:

Apple quietly knocking out beastly backend tech:


We can't fathom what is coming in the next three years:

Dennis with the facts. I'm generally accelerationist (which personally is more about being against deceleration rather than hell-bent acceleration). But many of us can't really grasp what is coming (see above). Dennis does understand what is coming and reminds us we have to take sincere care:

A framework for thinking about how to develop your AI ideas:

I've seen lots of clips from Jack in the past week from the same talk. His points here are why I'm a fan of jailbreaks. There's power in truth. The struggle between truth and safety is really the battle of this age:

Thanks for reading and send your feedback, I'm happy to keep improving this if you want something to change.

Subscribe to Tom Davenport

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.