Georgios Bakirtzis

Girl in metro split in two from Perfect Blue.

It's Kon for me as well.

Thanks, appreciate you taking into account my opinion :) Yes I am using freedom mode. I will take a look when time permits.

For me personally discussing about gaps allows students to be too vague (want to solve everything at once). So I'd rather them (and me) focus on a particular problem, of course part of that is hypothesizing but I am not sure I like the current way it is being presented wrt gaps.

The standard advice in my field is to answer the following questions: what's the problem? why is it an important problem (in a particular community)? what is your solution? what follows from your solution?

In the general setting I am in agreement with this advice: www.youtube.com/watch?v=vtIz...

I guess the standard advice is that when doing research you are speaking to a community, that community ostensibly has problems and you are trying to solve problems for them.

I tried this with one of my students. Mostly to make sure we are on the same page on things. I guess my biggest pet peeve is seeing science as having gaps rather than being in a continuum. I'd prefer an option to have a _problem construction_ instead of gap.

We’re particularly interested in researchers who care about 𝐦𝐮𝐥𝐭𝐢-𝐨𝐛𝐣𝐞𝐜𝐭𝐢𝐯𝐞 𝐩𝐫𝐞𝐟𝐞𝐫𝐞𝐧𝐜𝐞𝐬, 𝐳𝐞𝐫𝐨-𝐬𝐡𝐨𝐭 𝐭𝐫𝐚𝐧𝐬𝐟𝐞𝐫, and 𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐢𝐧 𝐝𝐞𝐞𝐩 𝐑𝐋.⁣
⁣
What you’ll do:⁣
• Design RL algorithms
• Conduct experiments of multiagent interactions⁣
• Develop theoretical tools for generalization and adaptation⁣

Institut Mines-Télécom - Post-doctoral fellow in model-based reinforcement learning - 24 months contract Who we are ?Télécom Paris, part of the IMT (Institut Mines-Télécom) and a founding member of the Institut Polytechnique de Paris, is one of France's top 5 general engineering schools.The mainspring of

Job alert: @ipparis.bsky.social is recruiting a fully funded postdoc in model-based RL. Apply by 𝐉𝐚𝐧 𝟏𝟎𝐭𝐡 𝟐𝟎𝟐𝟔 ⁣
⁣
To apply: institutminestelecom.recruitee.com/l/en/o/post-...
⁣
I’m happy to field informal questions via DM/email. Please reshare to reach folks who might be a great fit.

University of Virginia doesn't have any need for the state funding, it's a minuscule amount compared to the rest of the funding sources it gets. So the question becomes, what is the use of this letter?

No this helps. @gioele-zardini.bsky.social and I have been thinking about how to actually make category theory compelling and convincing and useful.

Thanks, appreciate the detailed view :)

And what is the thresholds that makes this usable compared to other applied category theory papers that are "mumbo jumbo"? Why did this one make the cut and the rest didn't for you?

Genuinely curious.

Interesting, there is some work that lexicographically prioritizes metrics for autonomous vehicles and I have done some of that. Is there something specific you do not to run into the "lexicographic dominance" problem, which would be a much bigger problem for LLMs?

In the south places like these come with more services, such as divorce attorneys and guns.

I've received my first unsolicited email from a crackpot that claims to have solved one of the millennium prize problems. I don't know how to feel about this.

Is this the beginning of the redneck militia?

That doesn't seem like a very useful response to my question.

Maybe I am not familiar enough with how this community measures things, but is "human PhD range" a useful metric? If so what does it mean?

I only wish they would also support org mode, but I do use it to share lists with people :-)

This is generally known to be true for expert writers: writers that are experts on a particular topic not experts at writing. See, for example, here www.youtube.com/watch?v=vtIz...

This is before LLMs: A student submitted verbatim code including the _other_ students name in the top comment. Failing the class, the student retakes and does exactly the same but with _another_ student. I was beyond flabbergasted.

This is the answer for me as well. The Iliad is clear on this.

I've never used HEY. I am curious to know what you think they do right? I have my qualms with how bloated the Fastmail UI is—and how it seems to be getting even worse in that respect—but it's overall usable, especially in comparison to say Gmail.

I really like Fastmail for personal email, mostly because of the masked email feature. I can create aliases for every vendor I use so there is no mail leakage and if things do go wrong I can immediately tell whose responsible. I would also consider it a really solid platform otherwise.

@j2bryson.bsky.social is the right person to ask.

That's a great resource. Maybe also: doi.org/10.1613/jair... as a complementary read.

I'm very interested in this type of work but I am confused about the "trick." Is it stitching experience from another trained policy? If so how does it differ from the initially trained one? I see the results but there could be multiple reasons the system is doing better in these metrics.