trq_

It's not an academic paper as far as I know, which is why I wanted to write this up. But the project certainly has a cult following (and cult opposition) on ML Twitter.

Appreciate the write up!

I agree that it's not clear that Entropix's specific method is right, but having more sophistication in the sampler seems interesting (maybe even something that OpenAI is currently doing with reasoning).

Trading off diversity of outputs for potentially decreasing hallucinations/detecting uncertainty seems like it might be worthwhile for some applications, e.g. agentic behavior. But definitely an open question, many evals needed.

Yes that's right, it seems like an area of more research.

Honestly it goes counter to the Bitter Lesson (http://www.incompleteideas.net/IncIdeas/BitterLesson.html, which stems from getting too fancy about maze traversal in Chess. But at the scale LLMs are at right now, the improvements might be worth it.

Yes that's a good point! To be honest, I felt that I wanted to try it on the machine I used every day, but it's definitely a bit risky. Let me link that in the article.

I guess I'm a bit naive, why is this?

What about if you are a founder with a minority stake? Would you recommend some other form of visa?

Hi Peter! How suited do you think the TN visa is for joining early stage startups? I feel like being in that space can involve switching companies more often than usual because of the failure risk. Is it risky to apply for a TN visa each time? Say you change jobs at most once a year.

Bethesda (fallout, elder scrolls) is basically the only AAA studio that makes an effort.

I'm mostly saying that AAA games not having modding is just a big loss, especially given how this generation of games (Fortnite, Overwatch, League) are magnitudes bigger than anything before it.

That's what was so great about editors like Warcraft 3, you get the AAA feel for free.

That depends on the effectiveness of the platform in integrating mods. DOTA was far more popular than normal Warcraft RTS games at its peak.

Fair but those are more for display/frontend purposes. They shutdown custom servers pretty hard.

If you're testing just how fun a game is, it's easier to get someone to try a Warcraft 3 'map' than install a whole new game, with a new interface, new art, etc to get used to.

I think the distribution networks and art assets still make it more effective than unity for testing game concepts. And that's what leads to games like Battle Royale and DOTA.

Yeah I think paid mods was off in execution not idea. Would love to see it come back.

The main thing that mods give developers is a distribution network and that is invaluable if you ask me. Getting someone to install your game is the biggest hurdle.

Your DLCs also expand the capabilities of the editors, because you need to add new functionality to program the levels hopefully (otherwise, you haven't really done much worth paying for).

Warcraft 3 Frozen Throne was an amazing hit, partly because it opened up the editor even more for content like DOTA.

That works when your mistake is on a machine and you fix it with code. When you make a mistake with people, apologizing is part of fixing it.

Ahh okay, try this? http://recordit.co/7TAT5qixzm I can't edit anymore.

Fixed, thank you!

PubPub | Product Designer | Boston, MA or REMOTE | Contract/Part-time

We're based within the MIT Media Lab, working on revolutionizing academic publishing. PubPub is a free and open tool for collaborative editing, instant publishing, continuous review, and grassroots journals. You can read more about the mission of PubPub at: https://www.pubpub.org/pub/hello.

We are looking for a product designer to join us for a few design sprints. For most of these problems we already have the feature built and overall design philosophy and ideas done, we are looking for help in designing the interactions and making them pixel perfect, as well as turning these design ideas into primitives that can be used consistently in our design. A general outline of our design guide can be found here: http://pubpub.github.io/pubpub-design-guide/

The ideal candidate thinks of themselves as a UI and UX designer comfortable working anywhere from wireframes to high fidelity mockups (CSS is a bonus). No coding experience is needed but the ability to work closely with developers and iterate on designs is key. Remote candidates are welcome!

An example of a target of a design sprint would be the 'header' data in a publication (a GIF of which is found here: http://g.recordit.co/7TAT5qixzm.gif).

Please send us a short introduction and demonstration of your work (behance profile, dribbble profile, PDFs, etc) to pubpub-careers@media.mit.edu!

PubPub | Product Designer | Boston, MA | Full-time (part-time possible) | On Site

We're based within the MIT Media Lab, working on revolutionizing academic publishing. PubPub is a free and open tool for collaborative editing, instant publishing, continuous review, and grassroots journals. You can read more about the mission of PubPub at: https://www.pubpub.org/pub/hello

We're looking for a versatile product designer that can think critically about users needs, design solutions and work at any point from wireframes to actual CSS. Any coding experience is a plus.

PubPub is built in a modern web stack using ES6, React, Redux and Node.

The position is based in Cambridge MA, we work out of the Media Lab and offer competitive salaries.

Please send us a short introduction and demonstration of your work (behance profile, dribbble profile, PDFs, etc) to pubpub-careers@media.mit.edu!

PubPub (www.pubpub.org) | Boston, MA | ONSITE | full-time

We're making a radically, transparent open academic publishing platform. People will use PubPub to write papers, but also to spin up independent, open-access scientific journals.

Currently based out of the MIT Media Lab, but paying competitively.

Looking for Javascript Devs and UX designers, more info here: https://www.pubpub.org/pub/jobs

Email pubpub-careers@media.mit.edu if interested!

Feel free to inspect the background page and look at the network requests we're making. We're doing requests to Google Analytics to store counts (e.g. how many installs we have, how many notifications are clicked, etc.) but there is absolutely no data from the notifications or any personally identifying information about you.

We're working on giving you a mute button. We want to make it _easier_ for you to manage your notifications.

Just launched 5 days ago! Over 100k notifications served.

We're completely client side, so there are no privacy concerns.