eugeneyan @eugeneyalt Twitter profile | Pikagi

Pikagi

eugeneyan

@eugeneyalt

652

Followers

72

Following

195

Media

1,408

Statuses

Care a lot, try hard, have fun. @eugeneyan 's inner Id.

https://t.co/Ui8IRoG7If

Joined March 2020

Don't wanna be here? Send us removal request.

Pinned Tweet

@eugeneyalt

eugeneyan

11 months

Note to self: • Work hard • Keep learning • Cherish loved ones • Find people who inspire you • Be kind & egoless • Eat healthy, exercise, sleep well • Read & write • Practice gratitude & meditate • Be present • Enjoy food & nature • Don’t sweat the small stuff • Smile

1

0

24

Last Seen Profiles

@MUNDOJUSTOYA

@naevistagram

@nishizawakeita

@vaguevamp

@lelocascio

@__xanushka

@omaigayt

@ALzimbabwe

@MarikaKorpinen

@NaoN290290

@thebriannamone

@kellyrusso21

@inside_the_rink

@Bifurca2

@CryptoPodroze

@Kelly_Tips00

@Chaitanya_World

@airgunweb

@mert_mp4

@DicksonWon3678

@sluglabs

@Gbgedhari1

@FinbotsAI

@notwisenough

@AsiwajuLahm

@alepand

@KeystoneSymp

@mariyankhan

@thevinayakjain

@NeuroAdvisor

@AsSa66986762

@bokeplokalmalam

@Drummdm1

@AsSa66986762

@thebriannamone

@DicksonWon3678

@eugeneyalt

eugeneyan

12 days

applies to most technical roles and prompted me to write this:

Tweet media one

@_xjdr

xjdr

12 days

AI is full of people that think what is difficult is valuable, lol

19

19

420

10

76

571

@eugeneyalt

eugeneyan

3 months

luck surface area > fly to SF to speak at conf > talk to 20+ pple > attend evals meetup > talk to 10+ pple > chief-of-staff hears me talk, texts friend to uber down asap > friend turns out to be research dir > we had highest alpha chat (1hr) on my entire trip > 3% hit rate

6

1

117

@eugeneyalt

eugeneyan

2 months

tbh it’s hard to articulate what i do to look around corners, connect the dots, force multiply, deliver through others. it’s especially challenging to explain how i do it or mentor others to do it too

@rakyll

Jaana Dogan ヤナドガン

2 months

As a senior individual contributor, part of how you get things done is always going to be black magic. In companies where there is an established senior IC population, this is a pretty well understood reality. In others, management will try to micromanage you or count some

11

53

644

4

5

60

@eugeneyalt

eugeneyan

3 months

look mum, i'm an "experienced LLM hacker"

Tweet media one

4

4

51

@eugeneyalt

eugeneyan

3 months

look who i bumped into the airport

Tweet media one

3

0

46

@eugeneyalt

eugeneyan

4 months

tech folks hired during peacetime ain’t ready for wartime

3

5

46

@eugeneyalt

eugeneyan

3 months

behind the scenes on how it started

Tweet media one

@eugeneyan

Eugene Yan

3 months

All three parts of "What We Learned from a Year of Building with LLMs" are now live on O'Reilly! • Tactics: • Ops: • Strategy: Read all 42 lessons here: A sample of what each

19

170

771

1

3

45

@eugeneyalt

eugeneyan

22 days

a lot of ml engineers and scientists have the misconception that “doing science” is about inventing or applying the sota, lots of math/code, etc. no, that’s not it. doing science is about valid measurements and experiments, understanding why it works or doesn’t, creativity, etc

0

7

46

@eugeneyalt

eugeneyan

4 months

look what Hamel did

Tweet media one

@HamelHusain

Hamel Husain

4 months

Love this essay from @eugeneyan This is especially acute for tools and infra around AI

Tweet media one

30

73

585

3

4

29

@eugeneyalt

eugeneyan

25 days

wow my paper on detecting hallucinations got accepted for oral presentation at Amazon ML Conf 🎉

3

0

28

@eugeneyalt

eugeneyan

3 months

them: "why don't you publish in a journal or conference? it'll help with your reputation" me: "do i really need to? feels like more people can benefit by sharing openly?" • • •

Start Here • Eugene Yan

Key themes I write about including machine learning systems & techniques, ML in production, data science methodology, writing, learning, and career.

@_xjdr

xjdr

3 months

man, i want to be like Noam when i grow up. This guy is out here doing native int8 training, local attention and radix tree kv_caching and then just posts about it casually on a blog. some of y'all would have milked 8 arxiv papers out of what was contained in that post.

11

15

437

1

5

22

@eugeneyalt

eugeneyan

3 months

@jxnlco

Tweet media one

1

0

22

@eugeneyalt

eugeneyan

1 month

drafting my 188th writeup (on llm-based evals) now and it's still incredibly hard to put thoughts on paper and organize it for my intended reader

Tweet media one

@eugeneyalt

eugeneyan

1 month

writing is HARD, and i've not been as productive this year as i'd like to be—only 8 write-ups & 2 talks😔 some of these took lots of effort: synthetic data, task-specific evals, what we learned, prompting, ml interviews. maybe I should write shorter pieces in larger quantity

Tweet media one

2

0

12

0

1

22

@eugeneyalt

eugeneyan

2 months

@sh_reya wait seriously? that's 30m input tokens?

Tweet media one

1

0

21

@eugeneyalt

eugeneyan

21 days

epic sf day trip catching up with old and making new friends. and amazing picnics, noodles, and jap dinner. 🤤

Tweet media one

Tweet media two

Tweet media three

Tweet media four

1

3

21

@eugeneyalt

eugeneyan

28 days

at a bbq someone asked why I "worked" every weekend. i said that for as long as i remember i always felt the need to catch up, from my psych degree to investment negotiations to data analyst to cs masters to ml eng to now llms. do it for a decade and it becomes habit

4

0

19

@eugeneyalt

eugeneyan

3 months

@willpienaar @sh_reya @ShreyaR @shrumm @devstein64 @tristanzajonc full workshop for @danielhanchen and @UnslothAI workshop at @aiDotEngineer

Tweet media one

2

4

18

@eugeneyalt

eugeneyan

3 months

the more you put out into the ether the more comes back at you applies to both good and bad vibes few truly understand and believe this

1

3

19

@eugeneyalt

eugeneyan

3 months

before this weekend i knew evals were important; after the weekend i realized it was a bottleneck for almost all other teams and it’s probably the most pressing issue for building reliable and usable ai product.

2

0

19

@eugeneyalt

eugeneyan

4 months

Top qns people get from L6 Applied Scientists (me): • What's the customer problem we're solving? • Do we have data/prototype that confirms this works? • Could we write it down so it's clear for everyone? • Have we considered latency, throughput, security, monitoring, cost?

1

2

19

@eugeneyalt

eugeneyan

1 month

working on my o1 and it's such a slog

7

0

18

@eugeneyalt

eugeneyan

4 months

@ekdnam hungry to learn & ship, ready to 996 for a deadline every now & then, pushing like runway's running out every day, when leader says jump they ask how high instead of bargaining to squat or jump later, willing to do whatever it takes ethically to serve users, and actually deliver

1

2

15

@eugeneyalt

eugeneyan

2 months

the inverse applies too: anyone who doesn’t have their own evals is just playing around

@yacineMTB

kache

2 months

@_xjdr anyone serious has their own private test bench that they hand wrote themselves

1

0

16

1

1

17

@eugeneyalt

eugeneyan

5 months

Just finished my draft of "A Builder's Guide to Evals for LLM-based Applications" If you have nothing better to do on this beautiful sunday and want to provide feedback on 5,000 words about evals for classification, summarization, and translation, copyright, and toxicity, dm me.

2

0

16

@eugeneyalt

eugeneyan

19 days

heartbreaking to see how poor leadership & management of a team / program leads to its obsolescence or hollowing out over years. it's not noticeable when you look at it month on month or quarter on quarter, but suddenly after a few years, its collapse is clear.

1

1

16

@eugeneyalt

eugeneyan

2 months

the last mile matters more than the first starting well but fumbling the last mile = poor outcomes and lost trust starting poor yet finishing well = solid outcomes and turning things around

1

3

14

@eugeneyalt

eugeneyan

3 months

dinner with some of my fav folks in SF

Tweet media one

4

0

15

@eugeneyalt

eugeneyan

3 months

overall, gained alpha from 6 - 8 folks i spoke with, so ~20% hit rate, with one outlier (3%)

0

0

14

@eugeneyalt

eugeneyan

2 months

@HamelHusain buying $AMD based on this Hamel

3

0

14

@eugeneyalt

eugeneyan

3 months

level 1: sign up for credits level 2: sign up for learning level 3: sign up for community level 4: sign up to forge friendships

@TheZachMueller

Zach Mueller

@TheZachMueller

3 months

*Signs up for a course* *Ends up making great friends out of @eugeneyan , @charles_irl , etc* I’d say it’s a win win all-around

0

0

18

2

2

13

@eugeneyalt

eugeneyan

3 months

eval notes • a few practitioners have methods similar to mine (simplify task, non-llm model-based evals, etc) and found it to perform well too • two experts: while they agree llm-eval not reliable enough now, confident can get there in 12 months and have invested heavily in it

2

0

13

@eugeneyalt

eugeneyan

12 days

the quality of your writing/coding tells me the quality of your thinking. if it's bloated, disorganized, or has redundant parts or large gaps, it suggests the same of your thinking. and if it's because there were no edits or rewrites, then it suggests lack of effort.

2

2

14

@eugeneyalt

eugeneyan

26 days

if you adopt a beginner's mind and accept you don't know everything and learn how others have succeeded; if you let go of ego and look beyond how things currently are, even if you contributed to it; you'll see a lot of bullshit that we should remove and lots of room to improve

1

1

13

@eugeneyalt

eugeneyan

3 months

senpai karpathy noticed ❤️

Tweet media one

@charles_irl

Charles 🎉 Frye

3 months

In this post, @eugeneyan , @BEBischof , @HamelHusain , @jxnlco , @sh_reya & I share our tactical tips for working with LLMs, from structured outputs to caching Stay tuned for two more posts covering the operational (hiring, product) & strategic (durability, competition) perspectives

Tweet media one

2

23

208

0

0

14

@eugeneyalt

eugeneyan

4 months

@HamelHusain wait a minute, it’s jAson and Hyaml 🤯

2

1

11

@eugeneyalt

eugeneyan

1 month

writing is HARD, and i've not been as productive this year as i'd like to be—only 8 write-ups & 2 talks😔 some of these took lots of effort: synthetic data, task-specific evals, what we learned, prompting, ml interviews. maybe I should write shorter pieces in larger quantity

Tweet media one

2

0

12

@eugeneyalt

eugeneyan

1 month

some non-ml folks ask me for a list papers to get into ml, and when I point them to applied-ml, they ask if i can suggest a handful to focus on. i don't think they understand: the list keeps growing; you need to keep reading and learning. I read 3 - 5 a week and barely keep up.

1

2

12

@eugeneyalt

eugeneyan

3 months

possibly useful resources • evals as system design pattern: • evals & hallucination for summarization: • practical evals for common tasks: • building a hallucination classifier:

1

0

11

@eugeneyalt

eugeneyan

4 months

company event; my table had 9 ladies + me brought cheesecake to table; she ate most of it♥️ she kept playing mobile games💕 she played scramble with friends. i started dueling her. after 2 weeks, got my chance: i formed RAMEN & asked if she liked it. first date. 12 yrs ago.

@coastalmama_

Mrs. G 🕊️

4 months

Please rt with the story of how you met your s/o. I want to know. Even if it’s “not interesting” 🫶🏼

Tweet media one

648

446

4K

1

0

12

@eugeneyalt

eugeneyan

16 days

wow when did i get a knowledge panel on google search

Tweet media one

1

0

12

@eugeneyalt

eugeneyan

2 months

it's tough competing with someone who's having fun

1

0

12

@eugeneyalt

eugeneyan

2 months

meta = zerg openi = terran anthropic = protoss

1

0

12

@eugeneyalt

eugeneyan

4 months

embarrassed to say i barely read books now. mostly been reading papers, writing notes, and sharing what i learn online (to reinforce my learning)

@rakyll

Jaana Dogan ヤナドガン

4 months

Industry wide AI burnout is real. People don't talk enough about it because they are too burned out to talk about it.

11

32

303

2

0

12

@eugeneyalt

eugeneyan

2 months

taking part in an ai hackathon, and wow, compared to this time last year, we've come a loooong way.

3

0

12

@eugeneyalt

eugeneyan

3 months

embrace the beginner's mind

Tweet media one

@AndrewArruda

andrew arruda 🎸

3 months

"your ability to keep doing interesting things is your willingness to be embarrassed again and go back to step one and start as a beginner and get your ass kicked and look stupid doing things." - @finkd

24

369

3K

1

2

10

@eugeneyalt

eugeneyan

2 months

skyline trail at mount rainier today 🥾🏔️

Tweet media one

Tweet media two

Tweet media three

Tweet media four

2

0

11

@eugeneyalt

eugeneyan

2 months

@HamelHusain saw that tweet; should have ignored and moved on. but i didn't, and now it's half hr past my usual gym time lol

2

0

11

@eugeneyalt

eugeneyan

4 months

this video made my day 🤣 punchline is pure alpha

Tweet media one

@granawkins

Grant♟️

4 months

sota RAG in 2024

64

233

1K

1

0

11

@eugeneyalt

eugeneyan

1 month

💯 working with LLMs over prolonged periods can make building / debugging data pipelines, conventional ml, recsys, and even doing on call, fun and something to look forward to lol

@_xjdr

xjdr

1 month

had a whole day of just normal ass API programming and slinging around protos like the old days. Turns out when you spend most of your time working with and programming for AI, regular old "deterministic programming" can be fun and kind of easy by comparison

5

1

112

1

0

11

@eugeneyalt

eugeneyan

2 months

> Further, it feels like base models are getting good enough that improvements to finetuning methodology don't matter much. never understood finetuning llamas and mistrals to improve on general benchmarks like mmlu or gsmk. feels a bit like kaggle?

@Euclaise_

Jade

2 months

I'm getting bored. Nous is amazing, but it's getting harder to find areas of overlap between my interests and theirs. Further, it feels like base models are getting good enough that improvements to finetuning methodology don't matter much. I'm not sure what to do

29

5

207

2

0

11

@eugeneyalt

eugeneyan

5 months

read this book years ago and figured out this is what I'll do

Tweet media one

@NirantK

Nirant

5 months

Reminder that the best thing you can do for your career is get so good that the best people find you and want to work with you Everything else: offer shopping, leetcode grindmax is less efficient

7

38

422

1

1

11

@eugeneyalt

eugeneyan

3 months

help i’d like to report a murder

Tweet media one

0

0

11

@eugeneyalt

eugeneyan

4 months

@jxnlco what are the requirements and why wasn’t this PM consulted? have you spoken to these 9 teams on how to reuse / integrate their systems? can you share the results of the user study research? did you get the approval of these VPs? if you want rate limit increase i need CEO approval

0

0

11

@eugeneyalt

eugeneyan

4 months

recently, i asked a mentor for advice, expecting them to tell me “North” to my surprise, they said the opposite, “South”, which was the harder, more uncomfortable path i acted on it; in hindsight it was right lesson: find mentors who shake you out of local optimal/comfort zone

2

2

10

@eugeneyalt

eugeneyan

4 months

working with people who have different values is tiring. in the end there'll be people who get taken advantage of.

1

0

10

@eugeneyalt

eugeneyan

1 month

i may have rediscovered the joy of lego

Tweet media one

Tweet media two

Tweet media three

Tweet media four

@eugeneyalt

eugeneyan

1 month

My Lego Star Wars X-Wing is here! Which would you do over the weekend? • Build the X-Wing • Work through backlog of papers on Model-as-Judge

4

0

1

0

1

10

@eugeneyalt

eugeneyan

5 months

@jxnlco > how do you encourage scientists or engineers to do the analysis/evals/prototype before implementation • make evals part of the product spec • make a baseline mandatory • share how simple always does better

2

0

6

@eugeneyalt

eugeneyan

3 months

@sh_reya Shreya, you gonna put me out of a job

0

0

9

@eugeneyalt

eugeneyan

2 months

if you say “let me know how i can contribute” while in a hackathon team, ngmi ☠️

2

0

10

@eugeneyalt

eugeneyan

6 months

@HanchungLee @lucy3_li 💯 prototype and iterate with LLM APIs till fit, then (maybe) scale with finetuning and self-hosting. the reverse often leads down a rabbit hole.

2

1

4

@eugeneyalt

eugeneyan

3 months

now i gotta rerun all my evals before tbe big gathering next week so i can seek advice more effectively lol

@AnthropicAI

Anthropic

3 months

Introducing Claude 3.5 Sonnet—our most intelligent model yet. This is the first release in our 3.5 model family. Sonnet now outperforms competitor models on key evaluations, at twice the speed of Claude 3 Opus and one-fifth the cost. Try it for free:

Tweet media one

448

2K

7K

0

0

10

@eugeneyalt

eugeneyan

5 months

• psych degree • internship: opened a patisserie, popular with 🌈 • govt: negotiated FTAs • ibm: only analyst without tech degree • e-com startup: no. 1 in southeast asia; acquired by alibaba • healthtech: predict & prevent chronic diseases • now: selling 📚 at a bookstore

@jxnlco

jason liu

5 months

Maybe I’ll write about my unconventional time at Waterloo. - physics lab - NYU epidemiology - action iq doing enterprise sales and forward deployed - stitchfix multimodal ai and search - facebook public safety and risk All by the time I was 22. - went to hackathons every

12

35

697

0

0

10

@eugeneyalt

eugeneyan

2 months

lol maybe we overdid it for evals

Tweet media one

2

0

9

@eugeneyalt

eugeneyan

3 months

@HamelHusain

@eugeneyalt

eugeneyan

3 months

just do your best and have fun; disregard critics and cheerleaders alike.

2

1

9

1

0

9

@eugeneyalt

eugeneyan

6 months

Call me thickheaded but I don't understand why a deeply researched, 40-min writeup on synthetic data gets less engagement than a 5-min hastily written post on unit testing 🫠

Tweet media one

5

0

8

@eugeneyalt

eugeneyan

1 month

every now and then i pause and reflect on my work and wow it’s crazy hard. the level of ambiguity, and how it extends beyond science and engineering into business and product and organization is 🫣 i guess best not to think too much about it lest i get paralyzed with fear 🤣🫠

1

0

8

@eugeneyalt

eugeneyan

3 months

just do your best and have fun; disregard critics and cheerleaders alike.

2

1

9

@eugeneyalt

eugeneyan

2 months

crap is this style of videos gonna do to ai engineers what it did to PMs?

@cto_junior

TDM (e/λ)

2 months

We're definitely in a bubble

Tweet media one

41

65

3K

2

0

9

@eugeneyalt

eugeneyan

3 months

what an exhilarating first half of the year. learned a lot, built a lot, shared a lot. gained several new friends, a few which will likely be 20-year friendships. kinda exhausted for now and can’t wait for my recharge next week 🪫

0

0

9

@eugeneyalt

eugeneyan

6 months

just built my own stitching together opus, vapi, and a vonage virtual number. took under an hour. it's a golden age for tinkerers and builders.

@swyx

smol swyx

6 months

I've now had multiple >20min phone calls with AI therapists and it feels completely natural. Every AI Engineer should be building their own therapist rn, and voice is the right medium. forget typing. go on a long walk and talk thru your day, your childhood, your dreams,

Tweet media one

65

49

771

3

2

8

@eugeneyalt

eugeneyan

3 months

@karthikkalyan90 many places i'd bet. the one i attended happen to be in sf. @aiDotEngineer also have an evals track at their conf

Tweet card media

AI Engineer World's Fair

Join 2,000 software engineers enhanced by and building with AI. June 25 - 27, 2024, San Francisco.

www.ai.engineer

1

0

9

@eugeneyalt

eugeneyan

3 months

@jxnlco the answer is in the map: "Toyota"

1

0

9

@eugeneyalt

eugeneyan

2 months

the difference between academic papers and industry write-ups is that the former’s finish line is reporting (loose) metrics like “54.9 spearman p” on static, overfitted data while the latter’s starting line is actually shipping systems for customer-facing experiences at scale

1

0

9

@eugeneyalt

eugeneyan

4 months

garlic butter with a side of shahi paneer

@gordic_aleksa

Aleksa Gordić 🍿🤖

4 months

ML interview question: what's your favorite nan?

12

0

13

3

0

8

@eugeneyalt

eugeneyan

3 months

i was hoping to learn from others how they solved evals. now i realized that everyone’s struggling too and i’m probably ahead of the curve. crap it’s gonna be tough.

@eugeneyalt

eugeneyan

3 months

before this weekend i knew evals were important; after the weekend i realized it was a bottleneck for almost all other teams and it’s probably the most pressing issue for building reliable and usable ai product.

2

0

19

0

0

9

@eugeneyalt

eugeneyan

3 months

And this is my favorite sentence in the writeup 🤣 Read it here:

Tweet media one

@charles_irl

Charles 🎉 Frye

3 months

My personal favorite in this section: "the rumors of RAG's demise are greatly exaggerated."

Tweet media one

4

9

100

1

1

9

@eugeneyalt

eugeneyan

3 months

looking at my nvda friends who tried to hire me and i declined 😅

@benln

Ben Lang

3 months

If you joined Nvidia 5 years ago as a mid-level product manager with an annual $70K stock grant over 4 years, just that initial grant would be worth ~$10.6M today.

133

1K

16K

2

0

8

@eugeneyalt

eugeneyan

16 days

for the first weekend in years, i don't have a topic i'm dying to write about, or an idea i really want to build so, at risk of coming across as arrogant, what's something you want me to write on or share? (alternatively, i'll revisit senpai k's class or build a labeling tool)

7

0

8

@eugeneyalt

eugeneyan

4 months

like how 10 harkonnen troopers don’t stand a chance against a single Freman

1

0

8

@eugeneyalt

eugeneyan

3 months

there are people who know how to talk about the thing, and there are people who know the thing—be the latter.

1

1

9

@eugeneyalt

eugeneyan

25 days

lgtm

Tweet media one

2

0

8

@eugeneyalt

eugeneyan

2 months

okay a third of y’all burnt out

@eugeneyalt

eugeneyan

2 months

i know several folks in my ai practitioner circle who are slightly burned out from learning, building, shipping ai products. quick poll: are you feeling burned out for AI?

1

0

0

2

0

8

@eugeneyalt

eugeneyan

3 months

there are people who're mostly talk and there are people who actually deliver—be the latter

2

0

8

@eugeneyalt

eugeneyan

4 months

llms raise the ceiling of expectations more than the floor of capability

3

0

8

@eugeneyalt

eugeneyan

4 months

💯 tech is the easy part; herding humans is much, much harder.

@shreyas

Shreyas Doshi

4 months

It is easy to make fun of PMs & engineers in big companies who are just launching features, moving pixels around, and are focused on “outputs over outcomes”, but you must understand that some environments are so messed up that a launch (really _any_ launch) is by itself a huge

27

96

716

0

0

8

@eugeneyalt

eugeneyan

3 months

interesting comparison between alt and main

Tweet media one

Tweet media two

4

0

8

@eugeneyalt

eugeneyan

2 months

you can’t change the people around you but you can change the people you’re around

1

0

8

@eugeneyalt

eugeneyan

4 months

interesting to learn where in the embedding space i’m in more influencers than i’d like though 🤣

Tweet media one

0

0

8

@eugeneyalt

eugeneyan

2 months

if agi was imminent, would K senpai be building a startup that teaches people how to build towards agi?

1

1

8

@eugeneyalt

eugeneyan

3 months

the size of a person is revealed by the size of what they consider problems

0

0

8

@eugeneyalt

eugeneyan

2 months

care for those who care a lot for love and camaraderie work with people who work hard to do your best work surround yourself with folks who have fun to enjoy life quarantine yourself from haters, karens, grifters, status seekers, freeloaders, drama mamas, negative nancies, etc

0

0

8

@eugeneyalt

eugeneyan

5 months

Me: <to my wife> "Wow look at this feedback" Wife: "Yea, I'm surprised people actually listen to three old men rambling on for hours with no agenda" 🫠

@eugeneyan

Eugene Yan

5 months

Did not expect this OH experiment to resonate so much. Thanks to everyone who's provided feedback, either via DMs or during the stream ♥️

Tweet media one

2

2

17

1

0

8

@eugeneyalt

eugeneyan

2 months

home sweet home. best sleep i’ve had in the week though still lots of sleep debt to catch up

Tweet media one

1

0

8

@eugeneyalt

eugeneyan

4 months

chatbots are such a pain to build esp if they have to respect the existing knowledge and act on it beyond just returning text

3

0

8

@eugeneyalt

eugeneyan

6 months

@HanchungLee @lucy3_li finetuning cost is usually a small fraction of inference cost, and inference of small classifiers is orders of magnitude cheaper and lower latency than medium sized LLMs

2

0

3

@eugeneyalt

eugeneyan

6 months

n years of experience may not mean the same thing for different people. for some it’s a solid foundation and virtuous cycle that accelerates their learning and future work. for others it’s outdated knowledge, bad habits, and hubris that holds them back and must be unlearned.

1

0

8

@eugeneyalt

eugeneyan

4 months

@jxnlco the strongest are the kindest

2

1

8

@eugeneyalt

eugeneyan

3 months

i just read the most amazing doc by an L7 that succeeded in saying a lot without saying anything. full of jargon and obviously written in a vacuum, divorced from data, analysis, or prototyping note to self: don’t write like that

1

0

8

@eugeneyalt

eugeneyan

4 months

Folks ask me why I don't just read a summary. > "Reading and writing notes on papers is how I learn."

@chrisalbon

Chris Albon

4 months

I don't want an AI to create notes on the new AI papers comings out. Reading and writing notes on papers is how I learn. I want an AI to find all the most important papers coming out and automatically add them to Zotero.

9

8

75

1

0

8