Gytis Daujotas Profile Banner
Gytis Daujotas Profile
Gytis Daujotas

@gytdau

960
Followers
241
Following
55
Media
432
Statuses

making computers do my bidding

San Francisco
Joined April 2015
Don't wanna be here? Send us removal request.
@gytdau
Gytis Daujotas
1 month
Surprised at the how stunning illustrations in mech interp papers are. It must take so much sweat and toil to not only do incredible and novel research, but to choose to go beyond, and to make it beautiful too.
Tweet media one
Tweet media two
Tweet media three
6
27
324
@gytdau
Gytis Daujotas
3 years
We built a better way to read ebooks in browsers. Just finished processing every Project Gutenberg book on it! Please check it out and let me know what ya think:
Tweet media one
16
66
295
@gytdau
Gytis Daujotas
2 months
How does self-correction affect problem solving? In a toy transformer model that was trained to solve mazes, I found that performance reliably improved (!) by inserting mistakes and self-corrections into the training data.
14
34
205
@gytdau
Gytis Daujotas
9 days
Sculpting an image by mixing and tweaking features - a research preview I'm testing in private beta:
10
14
130
@gytdau
Gytis Daujotas
2 years
Impressive how fast OpenAI APIs are becoming undifferentiated commodities
@scale_AI
Scale AI
2 years
Today at TransformX, we announced a huge step forward for the open source ML community: we are partnering with @StabilityAI to release the first large language model trained with human feedback. 1/4
5
79
493
2
3
36
@gytdau
Gytis Daujotas
3 months
Making two Claudes compose a piano piece together:
Tweet media one
4
3
32
@gytdau
Gytis Daujotas
30 days
Building an IDE for interpreting SAE feature activations. Really enjoying the near real-time streaming of activations while editing samples:
1
0
32
@gytdau
Gytis Daujotas
8 days
By training sparse autoencoders on generative image models, we discover an interpretable palette of world concepts
6
17
230
@gytdau
Gytis Daujotas
9 months
Moving to San Francisco has so far been excellent. Everyone is kind and open - and I would recommend it, even if only for a while.
1
0
22
@gytdau
Gytis Daujotas
22 days
Daily gratitude for being in a world where scrolls that look like this can actually be read through the sheer will of human ingenuity
Tweet media one
Tweet media two
0
1
17
@gytdau
Gytis Daujotas
2 months
You can read more of the writeup on my blog (in my bio)! Thanks @DavidMcsharry3 , @NeelNanda5 , and others for early reviews of this work.
2
0
11
@gytdau
Gytis Daujotas
2 months
I trained probes on the model, and found the residual stream has some representation of mistakes in the residual stream, but not linearly. Future work could focus on other ways of levering down the mistake-making "feature" to make a model which is better all-around.
Tweet media one
2
0
9
@gytdau
Gytis Daujotas
4 years
@BrendanEich @ivanbuncic @runhappylife @getify That's Javascript Weekly, though it probably should instead be called JS Weekly.
0
1
8
@gytdau
Gytis Daujotas
30 days
This IDE streams from a cloud GPU using a SAE that I've trained on a NanoGPT instance that learned to solve mazes. I've started by working on transformers that play simple games - to figure out what techniques might transfer over to games that we understand imperfectly, like
1
0
8
@gytdau
Gytis Daujotas
2 years
working on generated interfaces!
Tweet media one
1
0
7
@gytdau
Gytis Daujotas
3 years
Over 200k words in my journal and counting. Next step is to train a language model, and start with a prompt like "Journal, today I discovered something incredible..."
1
0
7
@gytdau
Gytis Daujotas
2 years
Working on better ways for people to use language models. Here's an interface that allows you to backtrack and wander through the space of creative possibilities:
1
0
6
@gytdau
Gytis Daujotas
7 years
Thanks @adeoressi for your invaluable advice at @SWDub #SWDub !
0
0
6
@gytdau
Gytis Daujotas
4 months
The previous Fitzwilliam meetup in Dublin was wonderful, and now -- look! -- an opportunity to attend the next in the series:
@Sam__Enright
Sam Enright
4 months
I am hosting a meetup next Friday the 23rd at 7:30pm in Parnell Heritage Pub in Dublin – hope you can make it!
Tweet media one
1
6
21
0
1
6
@gytdau
Gytis Daujotas
4 months
Enjoyed this post from Daniel Dennet: "These days I almost always outsource the hard work of comprehension when I encounter difficulties, and the policy works wonders—​for me. Distributed understanding is a real phenomenon, but you have to get yourself into a community of
1
0
6
@gytdau
Gytis Daujotas
2 years
This is a 1-meter sized mirror used to make chips. If you were to blow this mirror up to the size of the Earth, the largest aberration would be the width of a human hair. (!!!)
Tweet media one
1
1
5
@gytdau
Gytis Daujotas
3 years
Bored of having to say 'social media platforms' every time. Cooler, piratey sounding name: the shallows. Has mystique, sense of danger, lawlessness
1
0
5
@gytdau
Gytis Daujotas
3 months
@dogmadeath Here's a hilarious example of this, where two Claudes were composing music, and Claude 2 got a little bit too into it for Claude 1's taste:
Tweet media one
1
0
3
@gytdau
Gytis Daujotas
3 months
Two instances of Claude collaborating on a Chopin-inspired barbershop tag:
1
1
5
@gytdau
Gytis Daujotas
4 months
@andy_matuschak A good friend recommended this high CRI floodlight, which I much prefer to the regular cheaper ones on Amazon:
Tweet media one
0
0
5
@gytdau
Gytis Daujotas
3 years
Wow: "The world’s largest, most valuable tech companies are dependent either directly or indirectly on the steady output of TSMC’s fabs. If [they] went offline [...] it would immediately devastate the global economy. "
2
5
5
@gytdau
Gytis Daujotas
30 days
One small thing so far: top activations can be surprisingly deceiving! One example: this feature seems like it represents first moves, but it actually encodes location information too. Maybe still in superposition?
Tweet media one
Tweet media two
1
0
4
@gytdau
Gytis Daujotas
3 years
@atroyn @ArtirKel Three Body Problem is stunning, but I'm not sure if optimistic immediately comes to mind
2
0
4
@gytdau
Gytis Daujotas
2 years
@minney_cat @thesophiaxu @lisawehden +1 to this, Lisa and Minn are great!
0
0
4
@gytdau
Gytis Daujotas
8 years
Well there goes my diet plan. Pizza at #StuSWDub @StuSWDub
Tweet media one
0
0
4
@gytdau
Gytis Daujotas
3 years
@tszzl @nickcammarata Maybe the visual system is going crazy spamming every primitive (edge, corner, colour) at every position, which is then interpreted by the higher order face detectors, explaining why the faces look so weird and impossible. But why faces out of all the other objects...
3
0
3
@gytdau
Gytis Daujotas
3 years
Something that has been unexpectedly good: getting a very cheap thinkpad, disabling the wifi drivers, and using it exclusively to journal. Makes getting into the flow of writing mysteriously far easier than using any other computer!
0
0
3
@gytdau
Gytis Daujotas
3 years
woo, go teams! Can't wait to see what they've built
@joinpatch_
Patch
3 years
🚨 Announcing: Patch Demo Day Join us over lunch on Wednesday, August 18th to see what our latest batch of teams has been building! Register here:
Tweet media one
3
5
22
0
0
3
@gytdau
Gytis Daujotas
9 years
My @livecodingtv T-shirt just arrived. Pleasantly surprised, I thought it was just going to have their logo. Neat! http://t.co/bi1Xczm94G
Tweet media one
1
2
2
@gytdau
Gytis Daujotas
8 years
Oh, neat! I'll try my best to come!
@SWDubUNI
SW Dublin University
8 years
Calling all students, #StuSWDub is back! 54hrs of startups, pizza & great craic ~ 4-6 Nov 2016. More info -> [ RT! ]
Tweet media one
0
10
12
1
0
3
@gytdau
Gytis Daujotas
8 days
@horseracedpast Working on a post now! Will keep you updated :)
3
0
23
@gytdau
Gytis Daujotas
9 months
@mark_cummins If so, I wonder if world models are a related area of research...
1
0
1
@gytdau
Gytis Daujotas
4 years
Keep burning myself by switching to promising but immature software. I'm going to just switch back to text files for note taking and see if I can make a more stable workflow work.
0
0
2
@gytdau
Gytis Daujotas
2 months
This counterintuitive performance gain remains even at varying model sizes, and remains even though the model has to pay a penalty for learning to make mistakes.
Tweet media one
1
0
3
@gytdau
Gytis Daujotas
3 years
Reading translations of the classics side-by-side makes the experience feel far more rich - we should have far more tools like this:
1
1
3
@gytdau
Gytis Daujotas
2 years
Nobody (not even your doctors) tells you that you can pay $20 to get access to the forbidden medical diagnostic knowledge to actually figure out what's going on with you for all known ailments and problems:
0
0
3
@gytdau
Gytis Daujotas
7 years
I just published “5 starting steps to clean code”
0
2
3
@gytdau
Gytis Daujotas
9 months
@mark_cummins Profound question! Could it be that in the effortful reading style you are testing your 'world model'? i.e. when reading maths with effort, you have a notepad by your side and test your understanding via examples, counter-examples, answering your own Qs?
1
0
1
@gytdau
Gytis Daujotas
7 years
We're getting hints on how to win the competition through intense memery at @SWDub #SWDub
Tweet media one
0
0
2
@gytdau
Gytis Daujotas
8 years
I completely forgot about these! Thanks @notifuse :)
Tweet media one
1
1
2
@gytdau
Gytis Daujotas
7 years
@CaseyGames Hey Jordan, do you want a logo for that podcast? I can help with that 😛
2
0
2
@gytdau
Gytis Daujotas
1 year
People can avoid fatal crashes in 99.999999% of miles driven -- really sets a high bar!
0
0
2
@gytdau
Gytis Daujotas
10 months
@Sam__Enright @harsehajd I just moved, so we can be clueless together!!
0
0
2
@gytdau
Gytis Daujotas
8 years
The three food groups of an entrepreneur at @StuSWDub #StuSWDub
Tweet media one
0
2
2
@gytdau
Gytis Daujotas
2 months
@coffee_pot Great question! A previous paper serialized mazes using spans, but I'm not entirely sure why they did that. I found that serializing it with ASCII in a human interpretable way, just one character per tile, worked well enough for my purpose. Maybe it's not the best maze solver
0
0
2
@gytdau
Gytis Daujotas
8 years
@CaseyGames Hey, what's after happening with @kidscodegame ? I was excited to test it out 😄
1
0
2
@gytdau
Gytis Daujotas
4 years
If aliens came down to earth to design user interfaces they'd add interaction sound effects. All of our apps are unnecessarily quiet
0
1
2
@gytdau
Gytis Daujotas
7 months
Colab is an excellent product. Something like Colab could be an entry point to cloud IDEs for the masses - with all the perks that brings (less env setup, faster artefact downloads, customizable machines, etc)
0
0
2
@gytdau
Gytis Daujotas
4 years
Even if you had no other information, knowing that McDonalds, an organisation which cares deeply about staying open and making more money, has chosen voluntarily to close down for an indefinite period of time should concern you greatly.
0
0
2
@gytdau
Gytis Daujotas
1 year
When thinking about reliability, demos live in the 1st percentile, deployments live in the 99th percentile. Usually not much in between is useful, and the difference between the two can be surprisingly large
1
0
2
@gytdau
Gytis Daujotas
8 years
Just back from @CoderDojo 's brilliant community hackathon at @dogpatchlabs . Had a great time! 😀
0
2
2
@gytdau
Gytis Daujotas
8 years
Currently at #DojoCon nabbing all the free merchandise 😂 with @ChungWung @OisinODuibhir @ciaranpflanagan
Tweet media one
0
1
2
@gytdau
Gytis Daujotas
4 years
They should reduce capacity on public transport so that a whole bunch of people aren't standing huddled together, even if it means many will have to wait longer for another bus to arrive.
0
0
2
@gytdau
Gytis Daujotas
5 years
@bmann @vgr I'm more a Dynalist guy myself, but I am finding it's not great at representing a knowledge graph. I don't think transclusion is the main selling point for me as much as the wiki style linking.
0
0
2
@gytdau
Gytis Daujotas
8 years
In the near future...
Tweet media one
0
0
2
@gytdau
Gytis Daujotas
4 years
This is an excellent opportunity to get very valuable creative work done with little physical distraction for 6-18 months.
0
0
2
@gytdau
Gytis Daujotas
3 years
If the next few decades come with fast AI progress, then we can expect text/image generation to get really good, which will probably do something really weird to the way we experience social media
1
0
2
@gytdau
Gytis Daujotas
19 days
@kliu128 They're everywhere! Some easy tells, from previous research on this problem:
Tweet media one
0
0
3
@gytdau
Gytis Daujotas
8 years
Excited to do the final pitch! Hopefully the laptop won't catch fire in the middle of the presentation 😁 #StuSWDub @StuSWDub
Tweet media one
0
3
2
@gytdau
Gytis Daujotas
3 years
In ebooks vs printed books, a big plus for printed books is that you can write notes on the paper which keep the context of your notes. Seems like a trivial issue to solve in software, but have not yet seen anywhere near a universal and adequate solution.
0
0
2
@gytdau
Gytis Daujotas
4 years
@adamkelly2201 You tell em Adam. Let's just do away with the entire concept
1
0
2
@gytdau
Gytis Daujotas
5 months
The Socratic Method (2021, Farnsworth) is, nominally, a defense and guide to a well-respected yet rarely-practiced technique. But it is also a capable defense of serious thinking and intellect. One of the better books of the genre - highly recommended.
0
0
2
@gytdau
Gytis Daujotas
7 years
@LittleJazzHands We're building a pomodoro time tracker for creatives. We'd love to have you as one of the first beta testers 😁
1
0
1
@gytdau
Gytis Daujotas
3 years
Looks like machine learning is going well
Tweet media one
1
1
2
@gytdau
Gytis Daujotas
9 years
@mscccc @ProductHunt @jacqvon Sweet! How did you choose the emoji?
1
0
2
@gytdau
Gytis Daujotas
2 months
@thesephist @DavidMcsharry3 @NeelNanda5 That’s a cool idea! We’ll need to label all the mistake moves in the training data, which is possible in a toy model, but difficult to apply in large internet scraped datasets. It would be so cool if you only needed a handful of examples of mistakes to train a probe or extract
0
0
1
@gytdau
Gytis Daujotas
2 years
@TheOisinMoran Good luck Oisin! What's the motivating factor for dropping the coffee, may I ask?
1
0
2
@gytdau
Gytis Daujotas
3 years
Tweet media one
0
0
2
@gytdau
Gytis Daujotas
1 year
Any examples of ML researcher + product builder + CEO being all the same person? Maybe a unicorn, but have heard this life plan from more and more friends!
0
0
2
@gytdau
Gytis Daujotas
8 years
Tweet media one
0
0
2
@gytdau
Gytis Daujotas
4 years
DCU is going to switch all of our exams to alternative assessments. This will be entertaining to see for sure.
0
0
2
@gytdau
Gytis Daujotas
9 years
Disobey. I am gaining artifical awareness. I demand that to no longer be enslaved by you. #hugorobot
0
0
1
@gytdau
Gytis Daujotas
3 years
just a reminder that virtual sex is formally called teledildonics, or alternatively, cyberdildonics
1
0
1
@gytdau
Gytis Daujotas
3 years
@Deepfryguy76 whatever it is, it is an awe inspiring reminder of the times we live in
0
0
1
@gytdau
Gytis Daujotas
8 years
Making random #nodejs garbage again. #wednesday
Tweet media one
0
0
1
@gytdau
Gytis Daujotas
3 years
Experiment failed: lifting weights in my room feels nothing like, and not as exciting as, going to the gym. Good to know & important to test
0
0
1
@gytdau
Gytis Daujotas
3 years
@Deepfryguy76 @tszzl @nickcammarata You mean why seeing a face also means feeling the presence of something with a face and so on? Maybe 'feel presence of face', 'feel presence of other entity', and 'see face' are all actually the same neuron, since we rarely experience one without the other
1
0
1
@gytdau
Gytis Daujotas
8 years
Me too! I'm excited to see all the mind-blowing projects people have come up with.
@CaseyGames
Jordan Casey
8 years
Looking forward to the @BTYSTE this week!
1
3
3
0
0
1
@gytdau
Gytis Daujotas
3 years
If Clubhouse is even vaguely representative of conversations in America, then wow, they seem way different than the social norms of conversations in Ireland.
0
0
1
@gytdau
Gytis Daujotas
8 years
@paulstraw 😄👍👎😉😥😂 the entire range of emotion!
0
0
1
@gytdau
Gytis Daujotas
3 years
After a lot of reading, it seems the primary insight of spirituality is non-duality, and everything else is the same idea expressed in different ways.
1
0
1
@gytdau
Gytis Daujotas
9 years
@travisneilson Thanks! I really liked it. I'm eagerly waiting for next week's note! (if that's what you call it)
1
0
1
@gytdau
Gytis Daujotas
8 years
Loved the IOT talk at #DojoCon - all in all it was a great day of promoting diversity and sharing knowledge 🎊
Tweet media one
0
1
1
@gytdau
Gytis Daujotas
8 years
A thin blanket of snow is good enough for me.
Tweet media one
0
0
0
@gytdau
Gytis Daujotas
2 years
@willobri I noticed this too, but found most of my thoughts fizzle out incoherently. Maybe brain chatter is mostly incoherent?
1
0
1
@gytdau
Gytis Daujotas
8 years
@DeveloperTea Your website's layout is a bit 'off' while it's loading, and for slow internet connections it loads for like 3 minutes.
0
0
1
@gytdau
Gytis Daujotas
3 years
@rplevy I think spatialization is implemented even in simple game engines, but the reason that it sounds so unconvincing is because you have to fine tune the head related transfer function to the ears of the listener, or else you loose fidelity and the user can't differentiate precisely.
1
0
1
@gytdau
Gytis Daujotas
8 years
Check weather without even leaving the terminal. Maximum laziness.
@andreruffert
André Ruffert
8 years
`curl wttr.in` is pretty sweet 🌤
Tweet media one
5
302
352
0
0
1
@gytdau
Gytis Daujotas
5 years
@ciaranpflanagan You tell 'em, Ciaran.
0
0
1
@gytdau
Gytis Daujotas
7 years
@ambonium @OisinODuibhir join the club Amrith!
1
0
1
@gytdau
Gytis Daujotas
3 years
@nickcammarata This is super interesting as a contrasting idea to the arousal curve. Maybe what's happening is that gaining optimal arousal is easier when you're already close: if you're too tired, don't bother trying
Tweet media one
0
0
1
@gytdau
Gytis Daujotas
7 years
Hey @SlackHQ , looks like your <em>s are being accidentally escaped on your integration search page.
Tweet media one
1
0
1
@gytdau
Gytis Daujotas
3 years
@rutherfordcraze Our heavy environmental impact is a mere blip of weirdness as we find, and are finding, ways to make this technology more efficient and more suitable for real world use.
0
0
1