rue

20 somethin / sometimes researching / sometimes building / @getautocareer / a she

[rue @raw1dev]

introducing https://t.co/lzYalmnT7A

job hunting is legit broken

between babysitting chrome extensions, to spray and pray, spam apply, and enabling 100 applications a day per person, nothing is right

here's the thing:
1) nobody should be spending 12 hrs a week looking for a job simply bc of online application process
2) ai has become really useful and smart
3) quality matters just as much as quantity in a job search

to solve this, we made the flow work:
1.⁠⁠sign up, upload resume
2.⁠⁠browse as you normally would on LinkedIn. scroll, save, daydream
3.⁠⁠jobs you selected go into a queue for auto apply. agents work at your queue
4.⁠⁠each submission gets a confirmation screenshot and full web session replay
5.⁠⁠you get an email client live in the app. Every "Thank you for your application" and "Next Steps" and even security code emails are visible to you
6.⁠⁠the best part: an email, a session replay, an agent, etc. are all children to a job entity. so, when you get a callback to a job, the system will tell you what was filled in the form to submit, what resume was uploaded, what the cover letter said, etc.

we're in beta mode and looking for beta testers:
- 1M+ jobs worldwide
- 50 ATS systems supported
- 45,000 companies
- 20K+ jobs in last 24h

[rue @raw1dev]

@itsreallyvivek this is really cool. im employed and have publications in ai, its kinda awesome to see they accept a proper diversity. that being said i applied super late (rolling basis i guess)

[rue @raw1dev]

this assumption is so over the top

you should check OP's tweets and see what his standards of interview questions are considered 'acceptable' (it's some niche shit like what does this func in this niche lib do)

OP needs to accept that people naturally waste time, even at jobs they like and tasks they want to do, and learning on a daily basis in any job is rare. he's mega obsessed with python as a language for some reason, i respect it, and im that way too. but like, 99% of people coast a little bit

[rue @raw1dev]

@DevanshuXi seen u posting couple times on interviews, can u put them in a single thread pwease

[rue @raw1dev]

bad companies and bad roles too
not everyone that's actually smart gets placed into a good role even, depending on whatever the company needs and how they profile you and openings, you could literally be placed as some sales rep or code monkey (best case). it's very true in some unforgiving markets even if youre smart

[rue @raw1dev]

SOTA AI is not unimpressive to say the least (although the marvel really lies in data relationships and how well connected data is) .. it's just kinda the least impressive thing on the list tbh

[rue @raw1dev]

if you think about it it's a little bit insane that we designed, built and standardized things like computer chips, circuits, computer networking, assembly, high-level programming and infra before we invented math for vector matrices that predict the next token in a sentence

[Alberto Alfarano @albe_alfa]

Introducing Lattice Deduction Transformers: An 800k-parameter looped transformer that reasons like a SAT solver achieves 100% on Sudoku-Extreme with only 15 minutes of training.

A collaboration between @axiommathai, @AmherstCollege and @BarnardCollege.

[rue @raw1dev]

@xuanalogue IMO a lot of stuff needs to surfaced from pre LLM/transformer era

[rue @raw1dev]

@CathPoaster you know the video about the dotcom bubble where the guy is singing about all the details i want this in that format

[rue @raw1dev]

@jashvira @CathPoaster @darrowoah i swear earned my follow so fast

[rue @raw1dev]

@MichiganJustin same and this also applies to corporate. people can absolutely spend 1 hour+ in meetings saying a lot of things, i always struggled w this

[alphaXiv @askalphaxiv]

"Learn from your own latents, not tokens: A Sample Complexity Theory"

This paper explains why data2vec and JEPA can learn with much less data.

They showed that when data has hidden hierarchy, token prediction becomes harder as the hierarchy gets deeper. But latent prediction keeps the learning problem simple at every level.

Which suggests that models may learn faster when they stop predicting raw tokens and start predicting their own abstractions.

[rue @raw1dev]

completely wrong. genuinely curious are very few, most people can have that quality but it needs to be cultivated to show them that curiosity can allow them to work on something theyre interested in and most people never reach that path. the education system already teaches kids they're being trained for something instead of natively follow their curiosity to wherever it leads. curiosity is an extremely fragile thing that can die in youth if it doesnt lead somewhere fruitful and gets abandoned or it isnt cultivated

[Rohan Pandey @khoomeik]

my favorite interp researcher can identify neurons responsible for any behavior and provide steering vectors for them

her name is backprop and her steering vectors are just gradients

[Ben Dickson @bendee983]

Hot take: LLM tokens are a temporary phenomenon and will eventually be replaced.

Why? They are inefficient, expensive, and create all the wrong market incentives (e.g., tokenmaxxing).

Eventually, we'll have a more efficient means for AI systems to reason without the need to do it one token at a time (e.g., in latent space) while only producing enough text for humans to observe, control, and steer the model in the right direction.

[Harj Taggar @harjtaggar]

Chronic debaters are lethal to new ideas for the same reason pessimists are. They sound smart because they’re good at finding flaws before anything has had a chance to work.

[rue @raw1dev]

@varunram i'm pretty sure this is spam powered by ai agents?

[rue @raw1dev]

unpopular opinion but it's kinda embarrassing to develop and try to see what sticks with ai in public (coming from a technical or ai background) towards non tech people. as if it's already not embarrassing enough that big labs are constantly gloating that their models will automate everyone's jobs

[rue @raw1dev]

@JohnLocke504 @michalmalewicz i dont think hes talking *that small* like a founder who started a month back. it means startups that are well funded etc