
I've been using a chunk of my spare time to build out an agent runtime that is much more flexible than what exists today ime. Leviath is what I'm hoping will jumpstart the conversation around how adding structure to agents can improve their quality!


Focused on #AI #VR #AR #MR #XR #SpatialComputing | he/him | SWE @Meta | Previously @RealityLabs @amazon @alexa99 | Side projects @RGBSchemes | Opinions are mine

I've been using a chunk of my spare time to build out an agent runtime that is much more flexible than what exists today ime. Leviath is what I'm hoping will jumpstart the conversation around how adding structure to agents can improve their quality!


Looks like they removed it from the "cons" section, but have kept it in the rest of the article. This entire article is tainted with the focus on eye tracking throughout. To whoever @Gizmodo is doing these edits: Just removing the words "eye tracking" from the article won't fix this! This needs an entirely fresh review from someone who knows what they're doing, and can review a device while understanding that there is no eye tracking on it.

This post gets at the heart of what I’ve discovered working with generative UI: you have to give agents boundaries to make it functional. I’ll be sharing more very soon here. I’ve got some exciting stuff in progress.
I was wrong. Generative UI is not the future. My least favorite part about software is learning a new UI…and if new UI is frustrating then generative UI is just like a weird form of hell for users. I’ve though about this a lot btw, I literally made ArrowJS because I thought gen ui was an exciting future, but I’ve lost faith in it as a generic solution to software on-demand. Heck, even when you vibe code something, the worst part is figuring out the ui to know if it made what you wanted or not. I think generative UI should be tightly contained to “types”, like, on a dashboard where charts are, being able to say “make a chart of DAU over revenue” is perfect, now my expectations are pre-met.

If this ends up being true, I suspect the attach rate for such an accessory would be quite high. This could result in some interesting opportunities for devs to make new games that require it. Are we about to see a new generation of dual screen games? 😃
Nintendo Switch 2に着脱式の2画面目が登場か、DS・3DSタイトルへの対応に期待 sumahodigest.com/?p=76696

It's mind blowing how bad coverage of XR devices remains in mainstream media. @jamestpero over at @Gizmodo just posted a review of the Xreal Aura, and one of the "cons" is that "Hand and eye tracking are shaky sometimes" The device doesn't have eye tracking. I'm so over folks cosplaying as tech journalists. If you can't get the basic facts of a device right, why should anyone bother reading your takes? This review is completely irrelevant as a result.

Hot take: hobbies and especially having kids do far more for personal growth than hustle culture. Hustle culture tries to prioritize everything, resulting in sloppy decisions and bad products. Structured hobbies and family life force a person to decide what to prioritize and what not to, often resulting in stronger decisions and better results.
GFY Propaganda bullshit. Expected, I suppose, from a Nashi propagandist. I have 3 miniature humans that aren't my hobby - they're my entire raison d'être. To say nothing of my delightful wife that I greatly enjoy spending as much time with as possible. Every single late night at the office, business dinner, investor event/retreat, trip to a supplier, visit to a customer, etc is a direct trade-off against my being an involved and present father (and husband). It's a difficult balance! And yet I've done it. The real talk side of the "hustle culture" bullshit that nobody talks about is how your work output falls off a cliff around ~60 hours. Maybe a bit more if you're super young, but I doubt it. In most cases, that 80th hour at your desk is performative - it's theater. Balanced lives (whether it's raising kids, sailing, tinkering on hobby cars, fitness, whatever) directly lead to balanced thinking and a FAR more sustainable lifestyle than "all work no play." And, more importantly, it models for your employees that balance is actually good for output, morale, retention, etc. Building big, impactful companies is a marathon. An ultra marathon. 10 years of marathons back to back. Your mind (and body) do need to actually rest and restore, or else you burn out. Fact of life. Nonsense claims like this are propaganda from the (typically) VC crowd who want to see MOAR HOURS as if that makes PMF happen, under the misguided belief that hours in linearly equals company value out. Have hobbies. Have kids. Take a vacation with your partner. And also work hard to build your company. The balance is difficult to find, but very doable.

One of the biggest learnings I had a year ago now and has held up remarkably well: more code does not mean a better product This is very similar to how had managers and companies will often try to use amount of code as a measure of engineer quality. A lot of PMs in particular think that more code is a sign of more being done. This is of course not true, and the focus on this will kill any project exceptionality quickly. Many companies and people are going to learn this the hard way these next few years unfortunately.
this is whats going to happen to epic games in about a year

Also, it’s absolutely wild watching folks post these tools to GitHub and online and such. Nintendo’s lawyers are probably so excited for all the work they’re about to get on this. Easy pickings!

The comments on this are wild. The idea that piracy doesn’t have harm is so prevalent, and the consequences that are incoming from this are going to be wild. Most folks pirating are so short sighted in the impact this will have, it’s going to be a major wake up call. Modern gamers have never experienced a proper gaming recession like the 80s (myself included!) and I’m curious how the world will handle this.
Almost overnight, piracy has gone from 'we're making this unsupported triple A title available at low prices to counteract corporate greed' to 'I just wanna steal shit'. It's not about simping for a company, it's seeing how untenable this will be for an already fragile industry.

Discontinuing both the Cybertruck and Model X is extremely bad from a consumer trust standpoint. I say this as both a Model Y owner and someone who was looking into the Model X up until its discontinuation: how can I trust future Tesla products to not also be killed? This is Google levels of killing off products. FSD moving to subscription only, the slowing of building super charger stations, and the various promises over the years not being kept add to this feeling. At this stage, my next vehicle will be from a traditional car manufacturer or Rivian, simply because Tesla doesn’t offer what my family needs.
Tesla will be winding down production of the Cybertruck at Gigafactory Texas around the middle of next year to make room for increased Cybercab production

These solutions are all hacks on top of bad harnesses. The best solution is a harness where users and/or agents can control the priorities of what’s in the context window, allowing proper prioritization. IE: exactly what Leviath.dev does
Compaction is a mistake. When an agent's context fills up, tools like Codex and Claude Code compact it: 1. The model prefills the entire session again 2. It writes a summary of everything so far 3. A new session starts from that summary I measured this on a real ~400k-token session with Opus 5.5: → 38 s and $1.99 when the prompt cache has expired → 29 s and $0.17 when it hasn't And the summary is lossy. The exact error message, the file path, the one number you needed is often gone. So I do something new for Mainplane Context retrieval works the other way around. Mainplane keeps the full record of every session as a file. When the context fills up, a new session starts right away with one instruction: "You are continuing session X. Retrieve key context from it as you go." The agent searches the old session and copies the exact lines it needs, when it needs them. Same session, same model: 15 s and $0.10 to get back to work. It does not depend on a cache hit, and nothing is paraphrased. This is the bitter lesson applied to agent memory. Don't hand-design what the model should remember. Give it the full record and good tools, and let it decide. Mainplane is open source (MIT). Link in the first comment.

Curious about the story behind Community Notes? Tomorrow, its co-creator @jaybaxter joins us to share how a small experiment at X grew into a system fact-checking world leaders 10 AM PT tomorrow (Oct 30th) bring your questions: luma.com/2ijkp0nx

Hey folks, like a lot of people this year I've been affected by layoffs, and I'm looking for my next opportunity! I'm a Lead Unity Developer looking for a similar position. I have a decade of experience, mostly XR but in small and large teams, medical, simulation, etc. DM me!

Congrats to the team! Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!

Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.

somehow the corposlops refused to use my video