Jungwon @jungofthewon Twitter profile

Pinned Tweet

Jungwon

3 months

Looking for a Head of Design, Sr Product designer, Front-end & Full-stack engineers to help answer these open design questions as we build @elicitorg for scale!

1

3

26

Last Seen Profiles

@SLU_ForestManag

@zxltrxn_

@LoDmutBeShabat

@8nLE0iNun8Ro5wn

@yasinctnky34

@el_flako_5003

@fathichebbi

@MediaUnifor

@TatiNils

@ahmad_armaghan1

@EspenThoresen

@MOMO53KOUKAN

@AzraelAgain

@EGK7777

@6lgxuAUjv7BdHWw

@Robert01893457

@jkdoyon

@McwhirtFond

@crushyapapao

@Martoshouwu

@bakemono_tantei

@morye_teza_90

@DCipango

@_ORSSA

@GoldenDaveMno

@CoachLammie

@achlyssia

@06mkpeker

@VIPikuma

@illitwice

@Mirrortech

@EraseDictators

@not_mavfps

@ZhouGuanyu24

@DierdreGal8465

@ShurleyDaDon

Jungwon

@jungofthewon

8 months

the tow truck only fit one person. but we were two. the tow guy we called said he would come in his car to take both of us home. as soon as we were safely on our way, we asked him how many times he's had to rescue people from the side of the highway. this was a first! he

13

22

573

Jungwon

@jungofthewon

1 year

We’re “pivoting” Elicit with GPT-4 😉 Elicit in 2022 took unstructured text in papers and structured it into a table. Elicit in 2023 will take this structured text and enable you to “pivot” it, grouping it by concepts. Sign up here:

21

86

504

Jungwon

@jungofthewon

4 years

one of the best things to come out of @oughtinc hackathon today: @manda_ngo teaching gpt3 to reframe negative thoughts positively. cc @elicitorg

10

102

482

Jungwon

@jungofthewon

8 months

well just had the craziest, most bay area experience of my life. after dinner with the team, we were driving home in our electric car when, in the middle of the highway, the battery completely died (was at 60%). we barely had time to pull over before the car stopped and wouldnt

67

33

441

Jungwon

@jungofthewon

2 years

I'm much less excited about agentic language models and more excited about narrower LMs that are so cheap and fast I can run 1000 calls on every paragraph of text that exists in the world. I want to extract every claim in every blog post and paper published. Then run a full

8

34

433

Jungwon

@jungofthewon

2 years

How @elicitorg applies language models compositionally to make it easier to check models' work: 1 When language models extract info from or answer questions about papers in Elicit, users can quickly see the source - the part of the paper the model got its answer from.

7

38

318

Jungwon

@jungofthewon

2 years

My Twitter feed has gone from interesting ai content to annoying ai hype

13

19

274

Jungwon

@jungofthewon

2 years

1/ Everything we’ve built & learned decomposing reasoning tasks with language models!! We wanted to see if decomposition: - Helps LMs perform important tasks that are hard to train end-to-end - Improves troubleshooting and evaluation

Iterated Decomposition: Improving Science Q&A by Supervising...

Language models (LMs) can perform complex reasoning either end-to-end, with hidden latent state, or compositionally, with transparent intermediate state. Composition offers benefits for...

arxiv.org

8

35

215

Jungwon

@jungofthewon

8 months

the company is if you ever find yourself on the side of a highway in east bay and wanna get out

8

10

217

Jungwon

@jungofthewon

4 months

Motherfucking hero

Jan Leike

@janleike

4 months

To all OpenAI employees, I want to say: Learn to feel the AGI. Act with the gravitas appropriate for what you're building. I believe you can "ship" the cultural change that's needed. I am counting on you. The world is counting on you. :openai-heart:

242

415

5K

1

176

Jungwon

@jungofthewon

6 months

4 months after launching subscriptions, we've hit $1MM ARR :) Kudos @stuhlmueller @james_elicit @itsmesarahp @k1bird @BenRachbach @Mappletons @VivaLaPanda_ @LukeStebbing @__Charlie_G @OrangJulie43231 @etiennefd @tanay Greater things have yet to come.

15

10

152

Jungwon

@jungofthewon

2 years

Today LMs are wrong and make stuff up. But that will get better. Still, it's not enough for LMs to not say wrong things. LMs can read & write way more than a person can. They should tell me the absolute best, most right things possible. That's what we're aiming for @elicitorg

7

13

144

Jungwon

@jungofthewon

2 years

anyone can query and build on top of. I want The Human Genome Project" for text, especially academic text.

4

138

Jungwon

@jungofthewon

2 years

We've been prototyping a version of @elicitorg that pivots the table of papers into a list of "entities" across papers e.g. lists of techniques, datasets, side effects. How exciting is this? Would you prefer this to the table of papers in Elicit? How much would you pay for it?

10

9

103

Jungwon

@jungofthewon

2 years

systematic review analyzing thousands of papers for each claim to see how legit it is. I want to rewrite every paragraph from multiple different perspectives and biases. I want all of this to live in a dynamic, easy to use database that

2

1

102

Jungwon

@jungofthewon

4 months

Design is going to play one of the most important roles in the development of AI systems. It will impact how we interact with this raw intelligence, how the intelligence gets deployed, how it is evaluated, and how much we can trust it. It's an incredibly exciting time to

Adam Wiggins

@_adamwiggins_

4 months

I just finished my first article for @elicitorg as their “internal journalist” Aiming to publicly document the impressive work this team is doing!

9

27

191

3

15

98

Jungwon

@jungofthewon

8 months

the police came and we watched them slow down traffic. it was insane and i felt bad and very embarrassed but so so grateful for their rapid response and help.

4

0

82

Jungwon

@jungofthewon

2 years

AI today feels to me like covid in Jan 2020, or bitcoin in 2012. We're just on the cusp of some crazy stuff. How many of these will you live through before you decide to do something about it?

Benjamin Hilton

@benjamin_hilton

2 years

There are too few clear overviews of x-risk from AI. So I tried to write a new one. ( @80000hours ’ previous article was from 2017!) This was difficult - lots of articles on this are (imo) confused. I (hopefully!) did things differently: 🧵⤵️

3

23

137

4

5

85

Jungwon

@jungofthewon

9 months

someone recently described data visualization to me as "a way to see and understand phenomena that are impossible to directly experience because they are too vast across time or physical space" and i was deeply moved

1

11

68

Jungwon

@jungofthewon

8 months

grateful to be safe. grateful for public emergency and highway patrol systems that respond quickly and helpfully. grateful for private towing services and insurance companies. grateful to live in a place with such a high density of entrepreneurial people. grateful for founders

1

76

Jungwon

@jungofthewon

2 years

How to get fine-tuned GPT-3 Davinci performance at 10% of the cost

Charlie George

@__Charlie_G

2 years

1/ We've deployed a new model for abstract summaries in Elicit! We trained our new summarization model using reinforcement learning from AI feedback (RLAIF) similar to @AnthropicAI constitutional AI method. To try it, go to and type in a query!

29

69

422

0

10

70

Jungwon

@jungofthewon

3 years

AI safety via debate except one debater is a model fine-tuned on Eliezer's writings and the other is fine-tuned on Paul Christiano's writings.

3

1

64

Jungwon

@jungofthewon

18 days

"AI tools that actually work" is about the best possible compliment you can get as an AI tool right now

1

10

65

Jungwon

@jungofthewon

4 years

Everytime I listen to an @80000Hours podcast I'm shocked to find that I'm listening to @robertwiblin on 1x

5

1

59

Jungwon

@jungofthewon

4 months

39 + 2 and we out here

3

0

58

Jungwon

@jungofthewon

2 years

jamming on augmented reasoning prototypes & disney tunes with @thesephist @Mappletons @dmdohan paulbricman @oughtinc

3

56

Jungwon

@jungofthewon

3 months

Language models enable a ton of exciting new search techniques. There's a lot to do just with embeddings but one of the challenges with search is that everyone wants something different. Some people love the relevance-first focus of semantic search. They are blown away by good

2

11

56

Jungwon

@jungofthewon

5 months

Pretty cool to watch your runway get longer each month

8

1

53

Jungwon

@jungofthewon

2 years

4 Overall, I’m a fan of: - Keep LM tasks small, well-defined, easy to evaluate. - Once LMs do well at small tasks, roll up to bigger tasks. The compositional approach is better for eval & scaling. Small enough to evaluate. Easy to add more pieces. Why do it any other way??

3

1

50

Jungwon

@jungofthewon

3 years

We launched an early version of gpt-3 based classification in @elicitorg . Upload a spreadsheet with a few rows labeled. We'll tag the rest. e.g. classifying @foundersfund investments downloaded from @crunchbase into AI vs. not AI

1

9

50

Jungwon

@jungofthewon

3 years

@katgleason Looked through almost all 407 comments, including from various famous tech people and confirmed that there were only 2 mentions of @OpenAI which reminds me that people have no idea what's coming. what's already here.

4

1

50

Jungwon

@jungofthewon

6 months

Since the very first Elicit prototype, our team has tried to launch a user-facing feature every single week. This is a defining feature of how we operate. It is crucial to navigating a very open problem & design space. 2 years and 88+ features later, we've found a way to make it

1

5

49

Jungwon

@jungofthewon

3 years

Table 1: Impact of intermittent fasting on longevity Table view is now live in @elicitorg . Toggle to see each paper as a row and autoextracted population, intervention, and outcome for empirical studies. It's uncomfortably beta like all our launches! V0 ~ 60% accurate

2

11

48

Jungwon

@jungofthewon

2 years

So many computing legends were from the Midwest - from @WalterIsaacson 's The Innovators

2

11

44

Jungwon

@jungofthewon

2 years

We're looking for a product manager to build @elicitorg !

2

12

46

Jungwon

@jungofthewon

2 years

We first launched lit review in @elicitorg on Aug 31, 2021. 53 weeks have passed. We launched a user facing feature for 50 of those weeks. Often there were multiple features launched each week & lots of non-user facing features too. Incredible work, everyone @oughtinc .

2

7

44

Jungwon

@jungofthewon

2 months

New use case unlocked 🔓- Filter by author and get a quick summary of their work along with their papers Like @ylecun - what's he all about?

Elicit

@elicitorg

2 months

In addition to more relevant results, you can now run advanced searches using 14 filters. That means you can search based on: • Title • Journal • Author • Keywords • PubMed ID • DOI • Publication year • Open access status • Citation count And more. Here's an example

1

0

9

2

3

43

Jungwon

@jungofthewon

3 years

1/ Launched a new workflow in @elicitorg . With `Find experts` you give examples of the people you're looking for. We scale your list to 10s, 100s of people like your examples and find their Google Scholar & LinkedIn pages. Try it:

5

7

41

Jungwon

@jungofthewon

3 months

Tech culture encourages you to surround yourself with the smartest and most ambitious people possible. That is important. It is also important to surround yourself with the kindest and most virtuous people possible.

1

2

40

Jungwon

@jungofthewon

22 days

getting increasingly excited about utility agents that reliably perform some subtask (as opposed to more ambitious, splashy agents that unreliably attempt some major task). e.g. - agent that scours the internet for the pdf of a paper - agent that optimizes prompt for a specific

2

3

39

Jungwon

@jungofthewon

1 year

An analogy for thinking about moats for AI: In a world where any company can hire infinite numbers of smart "people" - do some companies win when others don't? Seems to me like still yes. What are those winning companies good at?

5

6

37

Jungwon

@jungofthewon

10 months

friendly reminder that someone can be fired from their job and still be a very good person! best not to conflate the two

1

38

Jungwon

@jungofthewon

5 months

Our team has been thinking about how to break down complex thinking into atomic steps that are easier to automate and evaluate for more than 5 years now ( @stuhlmueller , his whole life). With Notebooks we finally manifest that vision into a product experience that helps

Elicit

@elicitorg

5 months

Introducing Elicit Notebooks! Notebooks take your research broader & deeper, all on one page. You can now: - Run multiple queries on the same page - Combine papers across the Elicit database & your uploads - Combine papers across queries - Chat with papers - Select papers to

7

21

126

2

7

37

Jungwon

@jungofthewon

2 years

@peterwildeford did you guys pivot without telling me

2

36

Jungwon

@jungofthewon

11 months

Setting up for @elicitorg launch party @fiftyyears 🥳

Andreas Stuhlmüller

@stuhlmueller

11 months

Elicit launch party tomorrow night in SF! Our entire team will be there. DM me if you want to come

0

3

22

2

3

34

Jungwon

@jungofthewon

2 years

Elicit is about to become mind-blowingly elegant and intuitive. So so excited about @Mappletons pioneering new interfaces powered by language models.

Maggie Appleton

@Mappletons

2 years

Today was my last day @hashintel . I still support the team and everything they're working on ❤️ But I'm thrilled to be joining @oughtinc I wrote a bit about the move and Ought here: TLDR: Designing AI tools to help researchers with open-ended reasoning

19

7

149

3

1

33

Jungwon

@jungofthewon

5 months

while chat models has gotten a lot better recently, i still find that they fail to be comprehensive. sometimes i ask chatgpt for a list of things e.g. a bunch of regulations. it will give me a solid answer with examples. Then i'm like "well what about X?" and then it says: "yes

Jungwon

@jungofthewon

2 years

Today LMs are wrong and make stuff up. But that will get better. Still, it's not enough for LMs to not say wrong things. LMs can read & write way more than a person can. They should tell me the absolute best, most right things possible. That's what we're aiming for @elicitorg

7

13

144

4

3

33

Jungwon

@jungofthewon

4 years

Interactively brainstorming visions of the future with GPT3 in @elicitorg . Conveying visions that inspire and stretch the mind without being fluffy is super hard. I'm bullish on how much GPT3 will help people with this + scenario generation broadly.

1

4

32

Jungwon

@jungofthewon

1 year

Be a NYT real person. Use Elicit. 3 non-MSFT-OpenAI products are mentioned in this article. Elicit is one of them.

The New York Times

@nytimes

1 year

“It’s like collaborating with an alien.” AI has snaked its way into many people's everyday lives, while companies and scholars use it to supercharge work they could never have imagined. Here’s how 35 people are using AI for work, play and procrastination.

29

50

193

1

32

Jungwon

@jungofthewon

7 months

in just the last two weeks, i've talked to researchers working on next-gen sequencing, the gut microbiome, cancer epidemiology, cement manufacturing, sustainable aviation, bioinformatics, and microsilica about how elicit can help. research is so cool. there are so many people

0

2

31

Jungwon

@jungofthewon

10 months

feeling kinda bad asking chatgpt for things. it's probably super stressed and hasnt slept all weekend

2

0

30

Jungwon

@jungofthewon

2 years

Anyone want to try and use @elicitorg to make progress on the @alexeyguzey vs. @sleepdiplomat vs. Natália Mendonça sleep debate? Can pay you for your time. cc @Trent_STEMpunk

3

31

Jungwon

@jungofthewon

1 year

For the picks & shovels strategy to work with LMs, there needs to be actual gold in them hills. And the number of gold diggers needs to significantly exceed the number of pick providers.

4

0

30

Jungwon

@jungofthewon

1 year

need to be able to get language models from doing a cool thing once to doing it consistently 100 million + times at scale (and when the outputs of LMs are inputs of other LMs), small percentages of error rates really add up

2

29

Jungwon

@jungofthewon

3 years

Tell us if you want this and we can build it into @elicitorg

Andreas Stuhlmüller

@stuhlmueller

3 years

1. select text in browser, pdf reader, anywhere else 2. press command-option-enter 3. see hierarchical outline a la @RoamResearch @WorkFlowy

3

7

114

4

29

Jungwon

@jungofthewon

9 months

Seems inconsistent to dismiss ai safety claims as being based on scifi but then base ai hopes and dreams also on scifi

1

29

Jungwon

@jungofthewon

2 months

you're supposed to show them black & white pictures right

1

0

29

Jungwon

@jungofthewon

3 years

. @elicitorg gets to spend 15 minutes of fame on @newsycombinator homepage and i get to spend 15 minutes refreshing the page to see it move up and down

0

3

28

Jungwon

@jungofthewon

2 years

Approximately everything we know today about what @elicitorg will look like in the coming years. The roadmap goes from lit review --> many research workflows --> custom research workflows --> reasoning about evidence, arguments, plans & decisions beyond research

Andreas Stuhlmüller

@stuhlmueller

2 years

Our plan for Elicit

1

8

71

1

2

28

Jungwon

@jungofthewon

5 months

Since before Elicit & GPTs, our team believed that the way to scale up complex reasoning with AI systems was by decomposing reasoning & its evaluation into smaller tasks. Great to be back on the pod with @labenz , digging into the details of how this philosophy informs

Turpentine

@TurpentineMedia

5 months

In this episode of The Cognitive Revolution: @NathanLabenz sits down with @elicitorg co-founders @stuhlmueller and @jungofthewon to discuss their mission to make AI-assisted research more accessible and reliable. Learn about their unique approach to task decomposition, which

2

3

12

0

5

28

Jungwon

@jungofthewon

3 years

1/ Lots of interest lately in making language models “truthful”. How can we prevent GPT-3 from “lying”? We’ve worked on this in the context of @elicitorg . In Elicit, GPT-3 tries to answer your research question given abstracts from papers. (Can try at )

2

6

27

Jungwon

@jungofthewon

8 months

we called a towing service. they could come in 20 minutes. then we called highway patrol. we worried that the car lights would go out and this would be dangerous, especially given how far out into the highway the car was.

4

1

26

Jungwon

@jungofthewon

1 year

Now, with GPT-4, Elicit will also give you a table of concepts discussed *across multiple papers.* For each concept, you can see the discussion in each paper. Right now, it works best for finding benchmarks and datasets.

1

0

25

Jungwon

@jungofthewon

2 years

We've started streaming results as we compute them in @elicitorg , which makes everything run faster. Also, we're no longer limiting to 20 papers by default so you can review more papers. This is the groundwork for answering questions using the full text of papers!

1

2

25

Jungwon

@jungofthewon

4 years

i've been delegating more of my choices to a random # generator and it's working out great. examples: - should we finish the last mile of this hike - should i do #30 -50 on my todo list - for how many more minutes should i talk to this person - should i share this secret or nah

2

1

25

Jungwon

@jungofthewon

3 years

@RichardMCNgo tfw when you thought you were going to have an impact on the world via ai safety but it turned out you were destined for octopus advocacy

1

0

25

Jungwon

@jungofthewon

5 months

It's slightly terrifying in an exhilarating way to work with people who are so good they immediately force you to sit up straighter and level up. Super excited to have @_adamwiggins_ helping us navigate the balance between building a powerful general tool and deeply solving

Adam Wiggins

@_adamwiggins_

5 months

I'm starting a project with @elicitorg ! Elicit is an AI research assistant—like Google Scholar meets Perplexity.

8

11

129

1

25

Jungwon

@jungofthewon

2 years

Equity in exchange for compute. Compute is the new currency.

Anthropic

@AnthropicAI

2 years

We're excited to use Google Cloud to train our AI systems, including Claude!

21

105

933

0

2

25

Jungwon

@jungofthewon

1 month

An entirely new pricing model that supports everyone from high school students (researchers of the future!) all the way up to those pushing the boundaries of scientific sensemaking. Not easy to pull off given the diversity of our user base! But I hope this better supports the

Elicit

@elicitorg

1 month

🚨 We completely overhauled our pricing! We now offer an UNLIMITED FREE TIER that is the most generous of any research tool. Searches, summaries, and chats are completely free and completely unlimited, every single month. We also launched Elicit Pro, a new plan specifically

5

9

58

1

3

24

Jungwon

@jungofthewon

2 months

ngl repeatedly telling myself to be unburdened by what has been is going pretty well for me

0

24

Jungwon

@jungofthewon

10 months

never before have i felt like twitter is moving too slowly

0

24

Jungwon

@jungofthewon

2 years

"limitation of the study" is becoming my favorite custom column in @elicitorg

2

3

24

Jungwon

@jungofthewon

2 years

Thrilled to announce Illicit, the AI crime research assistant. Currently in private beta. also happy birthday @stuhlmueller

1

2

23

Jungwon

@jungofthewon

4 years

can't help but wonder if women would be more likely to test applications like this and maybe this is why representation in AI is important ¯\_(ツ)_/¯

2

0

22

Jungwon

@jungofthewon

1 year

We’re currently in closed beta for this prototype. Sign up here!! In the meantime, feel free to keep using

Elicit: The AI Research Assistant

Use AI to search, summarize, extract data from, and chat with over 125 million papers. Used by over 2 million researchers in academia and industry.

elicit.com

6

2

23

Jungwon

@jungofthewon

2 years

It seems like the key difference between “faster horse” and “car” is that “faster horse” is a result of anchoring on the solution whereas “car” is the result of focusing on the fundamental user problem.

0

23

Jungwon

@jungofthewon

2 months

We invest a lot in evaluations at Elicit. We want to be known for our accuracy and overcome the limited truthfulness of LLMs. So we’ve built a suite of tools to understand LLM quality for research tasks demanding high accuracy. We’ve tried every single option there is - hiring

Elicit

@elicitorg

2 months

To make sure Elicit delivers the most trustworthy research results, we must constantly evaluate it, and we must do that *at scale*. In this post @etiennefd describes how we evaluate LLM output, the tradeoffs between trust, scale, and flexibility and how we find our way around

7

30

0

2

23

Jungwon

@jungofthewon

2 years

some of the most counterintuitive things i've had to learn as a founder are the importance of getting out of the way, managing my energy & projecting positive energy, and ruthless focus. all of these often look like "working less." which is very awkward.

2

1

23

Jungwon

@jungofthewon

5 months

This reading list originally came together to help new people at Elicit learn the language model stuff they need. Surprisingly few (if any?) well-structured syllabi that go from basics (how does a transformer work) to most recent developments (zero-shot hyperparam transfer,

Andreas Stuhlmüller

@stuhlmueller

5 months

The @elicitorg ML reading list is now on Github with 140+ new papers LLaMA, Mamba, Mixtral, phi-1.5, zero-shot hyperparam transfer, weak-to-strong, tree of thoughts, lie detection, LMs for forecasting & ranking, GPQA, SWE-bench, calibration by pairs etc

2

11

74

0

4

22

Jungwon

@jungofthewon

2 years

2 Next - Elicit avoids collapsing complexity into scores. It decomposes things like "trustworthiness" into subquestions. It surfaces those subquestions to users, who can customize the way they want to evaluate trustworthiness in a given context

1

0

21

Jungwon

@jungofthewon

2 years

@seanjtaylor you tell me

1

0

22

Jungwon

@jungofthewon

2 months

This is the biggest improvement we've made to search in 2 years

Elicit

@elicitorg

2 months

We’ve completely overhauled Elicit’s semantic search! Quality has almost doubled in some cases - Elicit search results will now contain 78% more relevant papers, if you are searching for papers for a systematic review. Across all Elicit searches, you will find 40% more relevant

3

13

80

3

1

22

Jungwon

@jungofthewon

1 year

Ought's vision is to break down complex reasoning into parts. Panda will compose them together into beautiful symphonies. It was immediately obvious that Panda is a strategic, independent thinker. So grateful for this talented new teammate who is also delightful and human ❤️

Panda

@VivaLaPanda_

1 year

Just signed so I can finally say: I'll be joining @oughtinc working on infrastructure. Hyped to help take the amazing raw power of LLMs today and turn it into something that we can trust for use in high stakes contexts!

10

3

54

1

21

Jungwon

@jungofthewon

2 years

Progress in AI & crypto seem very closely linked to me. The more our world exists in bits, the easier it will be for AI systems to manage (control) it

9

0

22

Jungwon

@jungofthewon

4 months

Somebody buy @Mappletons a drink for having to wrangle all of these challenges on her own before we could even articulate them well enough to write them down

Maggie Appleton

@Mappletons

4 months

Adam is a brilliant writer and we're *very* lucky to have him working with us at @elicitorg . Did such a good job of articulating our complex UX challenges and tactics we've developed for designing with language models over the last ~2 years.

1

8

107

0

19

Jungwon

@jungofthewon

2 years

1.1 This is possible bc of how we produced that answer in the first place. First, we break the paper into small chunks. Then, we find the most relevant chunk. We give just that chunk to the LM to generate its answer. When users want to check, we show the chunk the LM saw.

2

0

20

Jungwon

@jungofthewon

11 months

@The_AI_Skeptic @JeremyPoley Losing $20 per user _on average_ considers the fact that some users cost them $80 a month

2

0

21

Jungwon

@jungofthewon

2 years

Thanks @chris_bail , @emilymbender , and others for prompting us to write this up. It's far from perfect but we hope it's a good start. Looking forward to your feedback.

Elicit

@elicitorg

2 years

Language model-based tools are new technologies with exciting potential but some current limitations. This is V0 of sharing how Elicit works & where it doesn't so that researchers know how to integrate it into their workflows and what to expect from it.

3

22

114

0

2

21

Jungwon

@jungofthewon

8 months

@gauravphoenix Not sure yet! The dealership says their ev mechanic doesn’t come in until Monday -_- Good thing we left all of our leftovers in the trunk

4

0

19

Jungwon

@jungofthewon

3 years

1/Results like this in @elicitorg are exhilarating. The sentence at the top is GPT-3 summarizing the abstract of the paper below to answer a question. Specifically, it synthesizes the first 3 sentences to produce a more elegant answer than any of the individual sentences.

1

3

20

Jungwon

@jungofthewon

3 years

Hopefully coming soon to an @elicitorg near you - convert all relevant papers into a row in a table. Automatically extract out key information across all papers e.g. study type, population, outcome measure.

Andreas Stuhlmüller

@stuhlmueller

3 years

Filtering by study population is my favorite so far. Here: me researching effects of l-carnitine @justsaysinmice

0

2

1

20

Jungwon

@jungofthewon

1 year

Progress in AI means intelligence will be as abundant as electricity. Most people still haven't fully internalized what this means. I think it basically means our world looks like this:

2

0

19

Jungwon

@jungofthewon

1 year

Made a super quick video of how i used @elicitorg to find a bunch of ML benchmarks. Still a lot of performance improvements to make but pretty sure there's nothing else like this? Also it's very annoying to have to do this manually

How to use Elicit for ML research

Use Elicit to find and organize machine learning papers, like finding relevant NLP benchmarks and prioritizing them along the criteria you care about! Try it...

www.youtube.com

1

2

20

Jungwon

@jungofthewon

1 year

it's crazy how much living & working in this world feels like fighting a war on my attention. it feels violent. things exploding around me, mind being attacked from all sides. meditation, prayer, mindfulness aren't passive. they're armors of steel. training for mental combat.

1

2

20

Jungwon

@jungofthewon

2 months

constantly trying to make @elicitorg more accurate

Elicit

@elicitorg

2 months

One of our primary goals is to make quality research faster while exceeding human levels of accuracy, so we're especially excited about some recent quality improvements to Elicit: - Data extraction in columns without high-accuracy enabled is now 15% more accurate - Using

1

14

0

20

Jungwon

@jungofthewon

9 months

Users starting to conduct their own systematic evaluations of @elicitorg vs. other tools! Lots of room to improve but at least Elicit doesn't hallucinate references.

1

7

19

Jungwon

@jungofthewon

5 months

Slowly catching up to the idea that restating what someone said often leads to as much new information as asking another very direct question (with better vibes)

3

0

19

Jungwon

@jungofthewon

1 year

i find myself saying "thank you!" to gpt4 after it's been helpful. is it just me

7

2

19

Jungwon

@jungofthewon

1 year

This is more than you ever wanted to know; it's overcommunicating how it works. Looking through the steps gives a taste of the superhuman scale at which LMs can operate. How long does it take us to read 20 papers, extract their key concepts, then group & organize them?

1

18