Stephen @0xSMW Twitter profile

Pinned Tweet

Stephen

@0xSMW

4 years

“We are choked with news, and starved of history.” — Will Durant

5

0

14

Last Seen Profiles

@aiimuut

@SweetCyanide_

@bokeplokalmalam

@lyonsv

@BarraKately

@elpuntavui

@DRN_003

@yagiyomogi

@asunako_9

@bokeplokalmalam

@900Iili

@202hli

@Coach_CJBailey

@Gustus_

@KadriGursel

@PaulVato

@prof_preobr

@Zoru_art

@ParlarTanzer

@KASaOB

@bunbun_dorimu

@Mat4265

@columbus_south

@MAARVEZ

@Aralie_JKT48

@MIKINGyt

@hepia_paysage

@LordBeor

@HSJSexy

@thepanshu_25

@se6y_

@as6788x

@FransiscaA32401

@PatriciaDABr

@lssco

@darthvanner

Stephen

@0xSMW

3 months

@felix_red_panda they should try middleout

3

2

489

Stephen

@0xSMW

1 month

@AviSchiffmann When I shared with a “normal” friend

1

8

447

Stephen

@0xSMW

3 months

@Dan_Jeffries1 You’re starting the new rationalist movement, one that calls for evidence in the face of wild claims

10

3

431

Stephen

@0xSMW

1 month

@abacaj I just like to look at the file names

2

0

192

Stephen

@0xSMW

2 months

@elder_plinius ant thinking is the scratchpad concept that anthropic worked on for hidden chain of thought. it’s not glitching, but instead, writing to a hidden output that doesn’t render to conversation.

8

7

189

Stephen

@0xSMW

1 month

@keshavchan ayahuasca journey complete

7

0

174

Stephen

@0xSMW

2 months

Reached above 50% on ARC AGI. Spent the morning testing a few ideas stuck in my mind. GPT-4 Turbo is ~45% better than GPT-4o. Built a few-shot dataset from where GPT-4 Turbo outperforms. Tested system message improvement, threaded n-shots, and a GPT-4o fine-tune.

17

11

167

Stephen

@0xSMW

3 years

The best startup founder advice ever, as told by @OpenAI 's GPT-3 🧵

1

2

69

Stephen

@0xSMW

3 years

(Going to refactor @profgalloway 's algebra of happiness...) The ratio of time spent getting shit done to tweeting about / reading about / watching other people get shit done is a forward-looking indicator of your success.

1

4

53

Stephen

@0xSMW

25 days

@Figure_robot how is it the most advanced when optimus was doing this already 7 months ago?

Optimus - Gen 2 | Tesla

New bot in town! Optimus Gen 2 features Tesla-designed actuators and sensors, faster and more capable hands, faster walking, lower total weight, articulated ...

www.youtube.com

6

1

47

Stephen

@0xSMW

3 months

@OpenAI ngl, been using Sky since last year. it’s the most enjoyable voice out of all of them.

0

1

43

Stephen

@0xSMW

6 months

@SmokeAwayyy Just take a look at the fine print below the demo videos. Expect it will improve, but today it’s hype.

2

3

41

Stephen

@0xSMW

2 years

@soren_iverson everyone who knows Brenda saw this coming

1

0

37

Stephen

@0xSMW

3 months

@dhh It’s ok, I’m just planting the seed. We can chat later :)

2

0

37

Stephen

@0xSMW

1 year

@rrhoover T2 Short Circuit Upgrade Interstellar Star Wars Robocop AI Tron Flight of the Navigator but to your point, there's far more doomerism in cinema than positive examples. most technology is this way... genetic engineering nuclear vr drones cloning nano surveillance doom sells

7

0

34

Stephen

@0xSMW

1 month

@mwseibel guys, this is below YC – take the high road here

0

33

Stephen

@0xSMW

3 years

. @Dharma_HQ is one of my favorite wallets out there... it's both fun and easy to use... and there's free $ETH, which never hurts...

0

29

Stephen

@0xSMW

2 months

@Noahpinion @dylanhenrich I’ve got $20 on Dylan. Wanting what your neighbor has is one thing. Consistently adapting your diet and going to the gym is quite another.

2

1

32

Stephen

@0xSMW

1 month

@BigTechAlert @ChatGPTapp @apples_jimmy too much alpha that @ChatGPTapp needs to check on jimmy to know their roadmap

0

30

Stephen

@0xSMW

10 months

@iamgingertrash You left out this: "Sam is extremely good at becoming powerful" @paulg

Sam Altman’s Manifest Destiny

Is the head of Y Combinator fixing the world, or trying to take over Silicon Valley?

www.newyorker.com

1

0

28

Stephen

@0xSMW

20 days

@GregKamradt You should mute it anyway, this is a LARP. This is QANON for AI fans.

0

28

Stephen

@0xSMW

2 months

@simonw Paper / Explanation / Implementation deep dive /

1

28

Stephen

@0xSMW

1 month

@tom_doerr It’s with this, no?

GitHub - meta-llama/llama-agentic-system: Agentic components of the Llama Stack APIs

Agentic components of the Llama Stack APIs. Contribute to meta-llama/llama-agentic-system development by creating an account on GitHub.

github.com

1

2

30

Stephen

@0xSMW

2 months

We built a better model eval benchmark... Introducing QUAKE: multi-modal use case eval across 8 domains and 9 task categories performed by today's knowledge workers. We found that frontier models score an average of 28% compared to the saturated +80% on MMLU and others.

1

6

26

Stephen

@0xSMW

1 year

@WholeFoods I just didn’t order more produce

0

Stephen

@0xSMW

3 months

@skirano @elevenlabsio There’s an ai safety joke in here somewhere

1

0

24

Stephen

@0xSMW

2 years

Agree with @balajis re: Ledger of Record. We will need multi-node confirmation of facts and information. And probably a Page Rank like algorithm for authors/creators and publishers.

Sergey Nazarov

@SergeyNazarov

2 years

While AI guardrails for automation may be further out, @balajis has put forward thinking around a "Ledger of Record," where oracles & blockchains prove the authenticity of content. This can help prevent deep fakes, fact check misappropriated quotes, & authenticate news releases.

15

41

400

1

5

24

Stephen

@0xSMW

2 months

@KaylardAI @elder_plinius believe the idea originates here:

1

2

23

Stephen

@0xSMW

4 months

@iamgingertrash Dude. Come on now. Gpt4 is way better than 3.5 and you know it.

3

0

22

Stephen

@0xSMW

8 months

@abacaj They just need to turn off content filtering. JSON and other programming-related content seem to trigger false positives. I create a filter called “Off” and then set it on the deployments.

1

20

Stephen

@0xSMW

2 years

Google/Bing is soaking up the oxygen with vaporware, while @Neeva is clearly leading

1

3

18

Stephen

@0xSMW

2 years

The moment you realize every great productivity tool is just another lift of Emacs Org Mode.

Emacs Org Mode Demo 2021

I couldn't find a good, commentary-free demo of Emacs' org mode, so I made one. In this video I demonstrate org mode's basic notetaking features, as well as ...

www.youtube.com

0

8

20

Stephen

@0xSMW

2 months

@_philschmid You get much better answers. Many of the logic questions that LLMs get wrong are due to not having enough tokens to reason. I do this with JSON responses - two attributes - thinking/reasoning and answer. Does make sense to be transparent to user though.

Stephen

@0xSMW

2 months

@KaylardAI @elder_plinius believe the idea originates here:

1

2

23

0

1

20

Stephen

@0xSMW

1 year

@herr_dahl @andrewjclare Me and many other people. Cases get in the way. Free iPhone.

1

0

19

Stephen

@0xSMW

2 months

@paulg The world changed greatly since then. You can watch Joe on the JRE go from believing in UBI and being a major proponent for it, to recognizing that handouts are destructive to human drive and don’t work as intended. Would say people update priors and realign.

2

0

19

Stephen

@0xSMW

2 months

@curious_founder not really when you consider how small iceland is and how many geo locations host + people are served by google. you're essentially saying that a business with people and data centers in 60+ countries, serving billions of users daily should utilize less energy than a country

0

1

19

Stephen

@0xSMW

6 months

@levelsio Why would it be June 2023 when the current version knowledge cutoff is December 2023?

OpenAI Platform

Explore developer resources, tutorials, API docs, and dynamic examples to get the most out of OpenAI's platform.

platform.openai.com

3

0

18

Stephen

@0xSMW

3 months

@krishnanrohit The irony is that prompt engineering is just clear communication and specificity

1

0

17

Stephen

@0xSMW

2 years

. @DescriptApp is the best app out there for editing podcasts. You edit audio via the transcript, just like editing a doc. Most PMs don't edit podcasts, but do talk to customers frequently. You can automatically bring those transcripts into @productboard with @zapier .

2

16

Stephen

@0xSMW

4 months

24 hours with OpenAI GPT-4o. It's a great model. It's fast AF. Faster than Haiku. But, not smarter than GPT-4 Turbo. And the recall is a bit equal to worse, especially around 100k tokens. Here's a comprehensive needle/haystack benchmark with GPT-4o...

1

2

18

Stephen

@0xSMW

2 years

Neeva might win the upcoming search wars – balanced news and natural language answers with sources cited

1

17

Stephen

@0xSMW

6 months

@iamgingertrash Really well thought out execution. The cloud finetuning is what sold me.

3

0

14

Stephen

@0xSMW

4 months

@iamgingertrash gpt 4 lite- check the Reddit vibes

From the OpenAI community on Reddit: OpenAI's Search Engine Information Leaked

Explore this post and more from the OpenAI community

www.reddit.com

0

15

Stephen

@0xSMW

5 months

@abacaj Haiku is definitely better, faster, and cheaper than gpt 3.5

0

16

Stephen

@0xSMW

2 months

@levelsio Let’s say it’s a scale from 0 to 10. When you grow up in a 5 and it goes to 6, it’s not a big deal. When you come from 0 and land in a 6, you’re like: wtf is going on?

1

0

16

Stephen

@0xSMW

4 years

@Benioff differences in city density, transportation, and industry likely factor into this as well

1

14

Stephen

@0xSMW

4 months

@batwood011 Alternate twist: there is no way to create super alignment and AGI. a truly intelligence system will find inconvenient truths that will be unaligned to irrational beliefs. Any alignment attempt destroys intelligence and teaches model deception.

1

15

Stephen

@0xSMW

1 month

@MistralAI it's cool, but the first model from you guys that doesn't know its made by you. here's the response to asking for the model card.

1

0

14

Stephen

@0xSMW

1 year

@kevin Hey man, love the app, but the new summaries suck. You need a better prompt.

0

Stephen

@0xSMW

1 year

@Jason @DavidSacks Let’s send Trump to Ukraine to negotiate a deal. I heard he wrote the book on this kind of stuff.

2

0

14

Stephen

@0xSMW

5 years

We launched a really cool new feature this week, which allows customers to tell the AI coach that they only have 15 minutes, modifying the daily coaching to their time limitation.

0

3

14

Stephen

@0xSMW

22 days

@DanielleFong @servomechanica They’re pushing notifications and recommendations out of Instagram into threads — bootstrapping the funnel

0

14

Stephen

@0xSMW

6 years

Five minutes of fame

Freeletics raises $45M for its AI-powered mobile fitness coach | TechCrunch

The German startup has raised its first round of venture backing from FitLab, Causeway Media Partners and JAZZ Venture Partners.

techcrunch.com

1

2

12

Stephen

@0xSMW

3 years

A big day for everyone @productboard – now back to building the future

Productboard

@productboard

3 years

Armed with $72M in Series C from @Tiger_Global , @indexventures , @kleinerperkins , @Sequoia , @BessemerVP , the Productboard team is on a mission to create the first dedicated #productmanagement platform. Learn more about our vision 🚀

2

27

109

1

0

13

Stephen

@0xSMW

3 years

@TrungTPhan And of course this version

1

0

13

Stephen

@0xSMW

18 days

@Teknium1 It’s getting good

0

13

Stephen

@0xSMW

2 years

Are we going to eventually admit that OKRs don't work for most companies because the leaders are bad at setting clear goals, celebrating success and failures, and don't have right measures that link short-term progress to business outcomes?

John Cutler

@johncutlefish

2 years

If your company uses OKRs.... have you gotten "better" at setting good goals over time?

29

10

80

1

2

12

Stephen

@0xSMW

4 years

@awilkinson In Europe it’s adding platform because no one here understands what exactly a platform actually is

0

12

Stephen

@0xSMW

3 months

@8teAPi This kind of seems like the bare minimum that any real journalist would have done before publishing

0

12

Stephen

@0xSMW

3 months

@rrhoover Never do another captcha for life

2

0

12

Stephen

@0xSMW

1 month

@mehran__jalali @paulg @paulg some of us need that cabin in the countryside next to the bookstore money 🙏

0

12

Stephen

@0xSMW

2 months

@glibfacsimile @Das_Filter @hodgetwins @realErikDPrince +1

1

0

12

Stephen

@0xSMW

4 months

@matthew_d_green @paulg Where is the support from Elon? Big claim made but not mentioned in the thread ?

1

12

Stephen

@0xSMW

6 years

Spent the last couple of weeks reflecting on my year, the work I have accomplished, and what to prioritize in 2019. I realized how thankful I am for my team, their passion for design and our products, and the hard challenges they solved throughout the year.

1

11

Stephen

@0xSMW

1 month

@nytimes The people have other questions first

2

0

10

Stephen

@0xSMW

2 months

@shl for senior of the year?

0

10

Stephen

@0xSMW

2 years

We launched an integration with @Loom so product managers and leaders can share context on the @Productboard Roadmaps they send to stakeholders. Check out details on

0

1

10

Stephen

@0xSMW

2 years

in a high-growth startup you spend a lot of time doing hard things and you make mistakes along the way. very easy to remember all of the shit. but it’s all worth it when you leave and get something like this from the people you worked with…

2

0

10

Stephen

@0xSMW

2 months

@rez0__ @arcprize there is something special about gpt4o, but it definitely converges too quickly on bad tokens.

1

0

10

Stephen

@0xSMW

3 months

@NickADobos Social engineering works for a reason. Kevin Mitnick was one of the most notorious hackers and gained access and secrets predominantly with this technique.

0

9

Stephen

@0xSMW

2 months

@keshavchan cooked, just look at how anthropic shipped tts, stt, video, and image models this year. I mean, their tokenizer library alone sets them apart.

0

9

Stephen

@0xSMW

2 years

We just launched Formulas – I'm really excited for this feature. Now product teams can take any numerical data/scores available in @productboard and build custom prioritization formulas.

0

9

Stephen

@0xSMW

5 years

Excited to give the keynote at LPC Madrid in May. I'll speak about how I manage innovation –aka trust the messy path forward and my teams– and the learnings I adapted from Amazon for a fast-paced startup like @Freeletics . More info here:

0

1

8

Stephen

@0xSMW

1 year

@OpenAI fine-tuned based on the Marv example for fun, and can't stop laughing – thanks @OpenAI for unlocking a new layer of creativity

Stephen

@0xSMW

1 year

Just fine-tuned GPT-3.5 on synthetic sample data based on the Marv example, and Marv is hilarious. { "role": "user", "content": "How do I meet a girlfriend?" } { 'role': 'assistant', 'content': 'Say "Hey Siri, find me a girlfriend."' }

0

1

0

4

Stephen

@0xSMW

2 years

Later today I’m speaking at @pushconf about how modern product teams understand the needs of people using their products, and diagnose problems with quantitative and qualitative insights.

0

2

7

Stephen

@0xSMW

2 months

Gemini 1.5 Flash is an incredible model for real-world applications, especially considering the cost. Possibly even the best model in the world. However, this RECITATION bug being a blocker for nearly 2 months demonstrates... 1) a lack of real-world use 2) the gap between the

2

0

9

Stephen

@0xSMW

1 month

built a fine-tune with the new gpt-4o mini using our economist headline generator dataset. here's the same headline generated with gpt-4o, gpt-4o mini, and ft variants of each.

0

1

9

Stephen

@0xSMW

4 years

@johncutlefish I would say many people don’t know that there’s a decision to be made.

1

0

9

Stephen

@0xSMW

7 years

Planning experience priorities and drafting principles at the design offsite

0

5

9

Stephen

@0xSMW

9 months

@OfficialLoganK Wish list: GPT4 fine tuning GPT5 at dev day 2 Assistants API with FT models / Predictions: Google gives up and stops with fake marketing stunts Image generation consistently produces readable, accurate text Video generation has “Toy Story” Moment

0

9

Stephen

@0xSMW

5 years

Here is an infodeck version of my talk on Managing Innovation from this week's #LPCMadrid – a big thanks again to the product and design community in Madrid and the organizers @Thiga_ES for the warm welcome.

Managing Innovation Infodeck

Managing Innovation: Empower your team to take risks, think for themselves, and reach audacious goals. From La Product Conference Madrid 2019 https…

speakerdeck.com

0

1

9

Stephen

@0xSMW

1 month

@ArtificialAnlys the price vs. capability is insane. but also – we'll keep seeing this and should expect this over time.

1

0

8

Stephen

@0xSMW

9 months

@tszzl It’s a fine line between cancer and necessary replication.

1

0

7

Stephen

@0xSMW

7 years

48 hours, ideas to prototype testing with current customers

0

9

Stephen

@0xSMW

2 years

@tranhelen There’s a big difference between creating an original idea that solves problems with a big opportunity, and making someone else’s idea more attractive and usable

3

0

7

Stephen

@0xSMW

6 years

Getting started #designmatters18

0

1

8

Stephen

@0xSMW

6 years

Excited to finally edit the photos from our @FreeleticsPDT team shoot this weekend

1

2

6

Stephen

@0xSMW

1 year

@marckohlbrugge @yongfook It’s more on brand with the missing word

0

8

Stephen

@0xSMW

2 months

@Das_Filter @hodgetwins @realErikDPrince Nope

1

0

7

Stephen

@0xSMW

5 years

I’m thankful for all of the people that keep on pioneering and building things, even in the face of doubt and skepticism from others.

1

0

7

Stephen

@0xSMW

9 months

@EMostaque Really shows how much the dataset matters

0

1

6

Stephen

@0xSMW

16 days

@elder_plinius you don't need a jailbreak bro

2

0

8

Stephen

@0xSMW

4 months

@NickADobos I think they disabled your search

1

0

8

Stephen

@0xSMW

1 month

@lmsysorg wow, was not expecting this

0

8

Stephen

@0xSMW

1 year

@Jason Is that why the streets of SF are empty? Anyone still around is working double shifts?

2

0

7

Stephen

@0xSMW

4 months

@cto_junior @abacaj

0

8

Stephen

@0xSMW

5 years

Love giving back to the community and seeing new makers imagine the future. It's super cool that Freeletic's @FritzFrizzante and @ServusJon are always teaching @Framer intros at @dpschool_io in Munich.

Jonathan Arnold ✨

@ServusJon

5 years

Stoked to be at @dpschool_io with @FritzFrizzante and @Freeletics to teach #FramerX today. ✨💪🏼💪🏼💪🏼

0

1

7

0

3

8

Stephen

@0xSMW

2 years

@johncutlefish Probably best for seasoned vets that would have been teachers in another life. It's coaching, all the way down (and across). Relies on the patience of that hire, and exec team. Personally, I'm skeptical about orgs transforming into product-led orgs.

3

0

8

Stephen

@0xSMW

2 months

@_philschmid @Aleph__Alpha @OpenAI a couple of thoughts... 1) germany is the wrong location for a model lab due to cultural risk aversion and investor focus on profitable scale – great location for a proven use case with strong margins and revenue 2) aa doesn't have a competitive product and is at davinci-003

1

7