anon.yaml Profile Banner
anon.yaml Profile
anon.yaml

@anonyaml

Followers
538
Following
1,013
Media
91
Statuses
1,881

yet another machine learning anon

Joined July 2023
Don't wanna be here? Send us removal request.
Explore trending content on Musk Viewer
Pinned Tweet
@anonyaml
anon.yaml
1 year
The shift in my life from everyone I know asking me to shut up about ML to everyone asking me how ChatGPT works has been so stressful. I need to pivot out of LLMs I can't handle the attention.
4
0
26
@anonyaml
anon.yaml
1 year
@QuiErat05 SMH people questioning this cmon we learned this in elementary school! Long hair=girl Short hair=boy!!
3
9
2K
@anonyaml
anon.yaml
10 months
wait LMFAO I just realized I know Roon in real life. I was his TA in undergrad
14
2
636
@anonyaml
anon.yaml
1 year
@QuiErat05 at the same angle with long hair it would be girl
2
0
440
@anonyaml
anon.yaml
10 months
he stopped showing up after a couple weeks since I was kind of a bad TA
2
0
228
@anonyaml
anon.yaml
10 months
@nearcyan @IsaacJLiu it will look about half as steep
2
0
222
@anonyaml
anon.yaml
10 months
He doesn't remember me 😭😭
@anonyaml
anon.yaml
10 months
wait LMFAO I just realized I know Roon in real life. I was his TA in undergrad
14
2
636
2
0
92
@anonyaml
anon.yaml
1 year
@JBasedos @RuneCodex @Trey_Explainer You're confusing irony with sincerity
0
1
64
@anonyaml
anon.yaml
1 year
@mrtimer2022 @alifarhat79 If their ability to change society for the negative is more powerful than your ability to change it for the positive who's really the "weak man" in this scenario?
1
0
63
@anonyaml
anon.yaml
10 months
0
0
64
@anonyaml
anon.yaml
10 months
@zebulgar @ryanseanbadger you're original statement boils down to: "I find it hard to respect people who are different from me" and it's not a very good quality
0
1
61
@anonyaml
anon.yaml
10 months
@satyanutella_ roons tweet is clearly ironic and mocking those speech pattern?
2
1
56
@anonyaml
anon.yaml
10 months
@mattparlmer When I Applied to ML jobs after undergrad: we only hire people with grad degree When I get a MS: MS is a strong no hire signal
2
0
55
@anonyaml
anon.yaml
1 year
@MindEnjoyer @MSJDpadawan Should a society ensure that 13 year olds can read? Is that tyrannical? The only 13 years old I ever met who couldn't read were the ones I knew in the homeschooling co-op that met in the church basement once a week and for whom that was their only learning for the week
2
0
48
@anonyaml
anon.yaml
10 months
@zebulgar @jazzplane the point is you don't know the situation of the people you are publicly stating you are biased against :/
1
1
54
@anonyaml
anon.yaml
9 months
I AINT STOPPIN!
Tweet media one
1
1
49
@anonyaml
anon.yaml
1 year
@coldhealing "no one talks about this" Everyone talks about it all the time I see people posting it every day
1
0
38
@anonyaml
anon.yaml
8 months
@krishnanrohit what is the x axis here?
4
0
33
@anonyaml
anon.yaml
10 months
@9thbeer I know right? I look much more like a tenured professor
0
0
35
@anonyaml
anon.yaml
1 year
@RuneCodex @Trey_Explainer It's a 50% chance. Either they did or they didn't. Not sure why everyone is so worked up about this
1
1
31
@anonyaml
anon.yaml
1 year
@MarkovMagnifico at 18 you can hook up with a random person at a party, get married, have a kid, get physically abused and divorce by 20. by 30 most of us have enough experience with people and relationships to select a partner based on compatibility and mutually aligned goals rather than lust
9
3
71
@anonyaml
anon.yaml
1 year
@ByRakeshSimha @marcthiessen @VivekGRamaswamy US economy can't sustain proxy war, but the much weaker Russian economy can sustain an actual war?
0
0
24
@anonyaml
anon.yaml
1 year
@mrtimer2022 @alifarhat79 Got it. The people controlling the future of the country are the weak ones. People off doing their own thing are the strong ones.
1
0
21
@anonyaml
anon.yaml
9 months
this is why I quit my FAANG job and now I'm working on a halting problem startup
@max_spero_
Max Spero
9 months
there is so much alpha working on a problem that everyone else thinks is impossible
1
0
6
4
3
24
@anonyaml
anon.yaml
10 months
@yacineMTB fyi when i say "rocks have qualia" i'm using it as engagement bait and a literary device to show how poorly defined my terms are. my honest opinion is that we need an epistemological advancement (basically Dennet is right). while the memes rn are dank, they're limited
3
1
24
@anonyaml
anon.yaml
10 months
@krishnanrohit the interest/dividends of 10M is much much more than my current expenses but really long term I want to buy a house, have kids, send to college, all of which might eat into that too much to maintain
2
0
22
@anonyaml
anon.yaml
10 months
@main_horse @_akhaliq some day we will look at BPE tokenization like we now look at bag of words
1
0
22
@anonyaml
anon.yaml
1 year
If you were assigned to work on an LLM latency related workstream and then realize that neither your manager nor tech lead know that transformer inference is quadratic what would you do?
4
0
21
@anonyaml
anon.yaml
10 months
@yacineMTB I think all JavaScript is blasphemy against the Holy Spirit
1
3
20
@anonyaml
anon.yaml
9 months
@abacaj they upgraded llama2 to gpt-35-turbo level really makes me wonder again what the actual size of 35-turbo really is
2
0
17
@anonyaml
anon.yaml
1 year
@svpino "You talk to people offline, and everyone is a die-hard TensorFlow fan." where are you going to find these people? I've still never met one at my job
2
0
16
@anonyaml
anon.yaml
10 months
@upstatefederlst finally someone got to the real heart of the issue, it's not a problem with the drug, it's a problem that women are prescribing it
1
0
16
@anonyaml
anon.yaml
10 months
@drkent247 @KneWKeeD @VivekGRamaswamy why should we treat humans as humans?
1
0
16
@anonyaml
anon.yaml
10 months
is it ok to write a paper that basically just goes point by point through another paper refuting it?
8
0
16
@anonyaml
anon.yaml
1 year
@rickyflowsinyou @a_musingcat That's a psyop propagated by Big Blank and Tan
1
0
12
@anonyaml
anon.yaml
1 year
Oh you're in "monk mode" huh? How many rosaries have you prayed this month? How many hours did you spend in adoration before the Blessed Sacrament? How many Bibles did you copy by hand? That's what I thought
0
1
13
@anonyaml
anon.yaml
1 year
For high throughput LLM apps with large static prompts before any user content, do you just bake the prompt into the model? Like freeze the memory state of the LLM after the prompt and then when user makes a query it can just go go go instead of reprocessing the prompt each time?
4
0
15
@anonyaml
anon.yaml
9 months
@rishmishra I think I muted bubble boi but blocked aella so I voted for the lesser of two evils with no other context
4
2
14
@anonyaml
anon.yaml
10 months
2
0
13
@anonyaml
anon.yaml
11 months
@Mzungu181 @quantian1 @mathphd_network If you burn the fat out of 3 lbs of waygu and eat it with ketchup then everyone involved loses :(
0
0
13
@anonyaml
anon.yaml
11 months
I always import numpy before pandas because I respect numpy a lot more. Every time I import pandas I get a little bit sad, and mad
2
1
12
@anonyaml
anon.yaml
1 year
@andrewChebyshev @taz_chu the fraction of papers with an undergrad first author is probably more than an order of magnitude higher in CS than in math
1
1
11
@anonyaml
anon.yaml
10 months
OpenAI models being 1-3 on a public blind preference leaderboard is worth 10x whatever the credit cost is for inference. Glad to see this project continue to be supported
Tweet media one
@lmarena_ai
lmarena.ai (formerly lmsys.org)
10 months
Just received, a big shoutout to @OpenAI 's incredible speed! gpt-4-turbo will be up again soon.
6
7
180
2
0
12
@anonyaml
anon.yaml
10 months
I'm in the gooncave, trying stuff. Some will work, some won't. But always gooning.
0
0
12
@anonyaml
anon.yaml
10 months
1
0
12
@anonyaml
anon.yaml
1 year
@JsonBasedman There are a lot of statisticians who unfortunately only know R. And some of them say things I'm interested in so alas I'm at least semi-lingual in R
3
0
11
@anonyaml
anon.yaml
1 year
@_dDeltaDt @zeta_globin eating healthy and exercising are like the best possible things you can do for yourself
0
1
11
@anonyaml
anon.yaml
11 months
Which Guillaume western man?
Tweet media one
Tweet media two
1
0
12
@anonyaml
anon.yaml
1 year
me to my data after the groupby:
Tweet media one
@andrew_n_carr
Andrew Carr (e/🤸)
1 year
This is exactly how I made zucchini gnocchi last night After I shixing the water and carefully into the agg, it was pretty easy
Tweet media one
6
9
108
1
1
10
@anonyaml
anon.yaml
11 months
Gemini pro is super trash at this stuff
Tweet media one
Tweet media two
0
0
9
@anonyaml
anon.yaml
1 year
@norabelrose LOL at benchmark performance guidelines. This community is extremely skilled at gaming benchmarks, the next generation will just add a fine tuning on 1299 level SAT scores no problem
2
0
10
@anonyaml
anon.yaml
1 year
@coldhealing Wtf I never knew Durant was talking to hot_bid??!! Durant gains +10 legacy points for that hot_bid is a legend
0
0
9
@anonyaml
anon.yaml
1 year
@knowclarified @NickADobos It could make calls and send texts. Which is what a majority of people did with phones in 2007
2
0
8
@anonyaml
anon.yaml
1 year
@bastard_brian "is there some hivemind we're tapped into?" Top name is the name of a recent Disney movie
0
0
8
@anonyaml
anon.yaml
10 months
@BasedBeffJezos Kardashev gradient is a great example of technically imprecise and incorrect language. Gives the vibe of cool progress but doesn't make sense since there's no indication that the Kardashev scale is differentiable or even continuous
1
0
5
@anonyaml
anon.yaml
10 months
@DanielleFong hate the left hate the right
0
0
8
@anonyaml
anon.yaml
1 year
1
0
8
@anonyaml
anon.yaml
10 months
@PlatonicThoth @tszzl best class in the whole Gooniversity
0
0
9
@anonyaml
anon.yaml
10 months
@creatine_cycle possibly one of the few places where the midwit meme actually accurately applies
0
0
8
@anonyaml
anon.yaml
11 months
every so often in a fit of hubris I get the idea that I can exploit some special case and implement something faster than the general version in numpy/scipy 100% failure rate but it's instructive every time
1
1
9
@anonyaml
anon.yaml
1 year
@TheSeaMouse @abacaj tokenizers literally ARE feature engineering
1
0
7
@anonyaml
anon.yaml
10 months
@creatine_cycle also between low IQ and high body count. Definitely not linear
7
0
6
@anonyaml
anon.yaml
1 year
@exitperfect what's the lesson? get as close to the source as possible and then just ignore what they say and do what you want instead anyway?
0
0
8
@anonyaml
anon.yaml
8 months
@ModerateMarcel @krishnanrohit I fucking love increments of 10, honestly a top 10 unit for me
0
0
7
@anonyaml
anon.yaml
1 year
Tweet media one
1
0
6
@anonyaml
anon.yaml
10 months
@omarsar0 seems irresponsible to write a review paper about Q* with no material to review
0
0
5
@anonyaml
anon.yaml
1 year
@yacineMTB @tohsin_ what we do in DL is already partial since we do the gradient of the loss wrt each parameter individually So while it is a function of many variables, the gradients are only wrt one, ie partial
1
0
7
@anonyaml
anon.yaml
1 year
The smartest thing to do is find the dumbest people in the world and copy everything they do! Because the dumbest and smartest positions are actually the same!!!!
Tweet media one
1
0
6
@anonyaml
anon.yaml
11 months
@minkbazink If this is the case, you're just not using the bleeding edge tech
1
0
7
@anonyaml
anon.yaml
1 year
@yacineMTB @ctjlewis I want to be able to pass different shit in in different circumstances and don't want to make 35 if statements and different functions to call in the different circumstances
2
0
7
@anonyaml
anon.yaml
1 year
@knowclarified @bryan_johnson It's determined by the lowest number a doctor will say after this guy continuously offers more and more money to say a lower number
0
0
7
@anonyaml
anon.yaml
1 year
If someone has a variant of "e/acc" in their bio that isn't "e/acc" it's an intentional move to signal to the e/accers to get support, while also getting the plausible deniability of "doing it ironically" to distance themselves from the cringe
2
0
6
@anonyaml
anon.yaml
10 months
@zhil_arf @zmkzmkz "we humbly offer a solution to the ... problem"
0
0
5
@anonyaml
anon.yaml
10 months
@Frozenfire42 @weirddalle neutral is certainly an interesting word choice for a person financially supporting one side of a conflict
0
1
7
@anonyaml
anon.yaml
1 year
@bambipotf @evanjconrad Yeah, attitudes like this are why people hate tech bros. The smug superiority to assert that they are the only people in the world trying to solve problems. Makes me embarrassed to even be in tech
0
0
7
@anonyaml
anon.yaml
11 months
@QuetzalPhoenix @evanxitter Wow, all those heritage sites really get the digestive tract going
0
1
6
@anonyaml
anon.yaml
10 months
@BasedNorthmathr if you believe the simulation hypothesis then you can imagine a situation where the simulation authors made a bug which we could hypothetically exploit
1
0
6
@anonyaml
anon.yaml
1 year
@yacineMTB this guy keeps on posting on twitter. He posts all the time. good for him, some of them are funny. but I don't get it. doesn't he have a basketball??
0
0
5
@anonyaml
anon.yaml
1 year
@Indian_Bronson @default_friend How many Druze prime ministers has Israel had?
0
0
6
@anonyaml
anon.yaml
11 months
@jxmnop is this really true though? Where is the information about the next token stored? In the final layer activations. Where is the information which produces an embedding? In the weights. Don't actually see the distinction here
0
0
7
@anonyaml
anon.yaml
11 months
@BasedBeffJezos If you don't talk to puppets why did you repeatedly request to be on his podcast? 🤔
0
1
7
@anonyaml
anon.yaml
8 months
twitter has turned into an endless scroll of people posting photos of AIs commenting on photos and the posters don't seem to have noticed that the comments are from AIs
@venturetwins
Justine Moore
8 months
Facebook has turned into an endless scroll of AI photos and the boomers don’t appear to have noticed
Tweet media one
Tweet media two
2K
5K
104K
1
0
7
@anonyaml
anon.yaml
1 year
Tweet media one
0
0
6
@anonyaml
anon.yaml
10 months
started writing a paper today, this year I'm finally gonna do it. Final confirmation that I needed to is seeing others papers working on the same issue and they are just bad and I already have better results
0
0
6
@anonyaml
anon.yaml
8 months
@NPCollapse actually we have no way of knowing how many blue cubes are represented in the picture
0
0
6
@anonyaml
anon.yaml
1 year
@yacineMTB If you figure out "how deep learning really works from first principles" let us know because so far nobody has gotten very far on that one.
1
1
6
@anonyaml
anon.yaml
1 year
I'm allowed to have 3 books in progress at a time: one sci fi, one philosophy/history/religion, and one other non fiction. Any more and progress on all of them goes to 0 because of attention split and anxiety
1
0
6
@anonyaml
anon.yaml
1 year
@Thedonofstocks @paulg If real estate is 1.5x and actual people being there is 1/3 it just means there is a real estate bubble
1
0
5
@anonyaml
anon.yaml
9 months
@tmdanis If you only lived on twitter you would think autistic just meant smart and based
0
0
6
@anonyaml
anon.yaml
1 year
@AdamNeumannsCoS Are you asking: "why would some who is great at something switch jobs?" Because someone else will pay them more to do it than their current company
1
0
4
@anonyaml
anon.yaml
1 year
Q: do LLMs (GPT4) take longer to process and generate the same number of tokens if you are asking it to do a difficult task vs an easy one?
Yes
29
No
29
11
0
4
@anonyaml
anon.yaml
10 months
@JsonBasedman because you're gonna do layer norm anyway so constant factor is kinda useless
2
0
6
@anonyaml
anon.yaml
10 months
@yacineMTB Does this work with penises?
2
0
6
@anonyaml
anon.yaml
8 months
If you release a model and don't even claim that it's better than gpt-4 is that bullish because you seem honest, or bearish because you couldn't even get close enough to COT @32 your numbers?
1
0
5
@anonyaml
anon.yaml
9 months
@vboykis I asked GPT-4: "can you please visualize attention better than this?"
Tweet media one
1
0
6
@anonyaml
anon.yaml
10 months
ok now let's you build it in survival mode
@philipturnerar
Philip Turner
10 months
21,264,880 atoms and 2,528 parts. This is nanofactory material. [1/19] #MNT #APM #nano
26
92
613
0
0
6
@anonyaml
anon.yaml
9 months
Yesterday I tweeted this as a joke. Today I remembered that in high school I devoted several hundred hours towards trying to find a counter example to the four color theorem.
@anonyaml
anon.yaml
9 months
this is why I quit my FAANG job and now I'm working on a halting problem startup
4
3
24
0
0
6
@anonyaml
anon.yaml
9 months
1. dynamic/online rating systems 2. linear algebra graph algorithm implementations in pyspark 3. hugo award winning novels 4. the documentary hypothesis and theories of biblical authorship 5. why the federal government should not be involved in party primaries
@jxmnop
jack morris
9 months
What are five topics you can talk about for 30 minutes with zero prep 1. information content of text embeddings 2. architectural mysteries of transformer language models 3. the meta-research process (how & why to pick problems) 4. whether the NBA is rigged 5. shrimp suffering
13
0
83
0
0
6
@anonyaml
anon.yaml
1 year
@yacineMTB on the contrary, if you are coming across leetcode style problems in your your projects you are working on surface level uninteresting projects and are ngmi sadly
1
0
6
@anonyaml
anon.yaml
10 months
@yacineMTB I've solved so many bugs in the shower, they really should have showers in every office to enable the highest level of problem solving thought
0
0
6