HumanLevelJen
@HumanLevelJen
Followers
119,412
Following
460
Media
108
Statuses
1,079
@Co -Founder / Chief AI Officer @KIPprotocol | Humble Functionary @daobitreasury
Singapore
Joined December 2012
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#IndependenceDay2024
• 320556 Tweets
台風接近
• 179036 Tweets
終戦の日
• 153484 Tweets
台風の影響
• 145790 Tweets
台風7号の接近
• 133132 Tweets
Jai Hind
• 116302 Tweets
終戦記念日
• 103109 Tweets
#स्वतंत्रता_दिवस
• 78181 Tweets
上手上手
• 48545 Tweets
लाल किले
• 46249 Tweets
Gena Rowlands
• 39663 Tweets
誕生日グッズ
• 34311 Tweets
開示請求
• 33254 Tweets
Floyd
• 30507 Tweets
taekook
• 25533 Tweets
メカクシ完了
• 19247 Tweets
Buon Ferragosto
• 17320 Tweets
ディズニープラス
• 16280 Tweets
カゲロウデイズ
• 14130 Tweets
日焼け止め
• 13956 Tweets
カゲプロ
• 12676 Tweets
臨時休業
• 11857 Tweets
Live now :)
Join us this week at
#MYBW2024
in Kuala Lumpur! 🇲🇾
Our CEO,
@Julian_kip
, will be speaking at the
@Offchainglobal
Global Conference 2024⚡🎤
📅 Date: July 30, 2024
🕒 Panel Time: 1:10 PM
If you're at the event, come say hi! And don’t miss our panel with other industry leaders
239
486
640
176
235
520
Full disclosure: I only have a couple of resale 4090s and thus have never been invited to the CNC orgies.
To the journalists contacting me about the AGI consensual non-consensual (cnc) sex parties—
During my twenties in Silicon Valley, I ran among elite tech/AI circles through the community house scene. I have seen some troubling things around social circles of early OpenAI
204
605
4K
7
84
134
HK Web3 Festival/HKU Big Data Studies Lab with
@julian_kip
(shout out
@javiercha
)
Expect big news soon...
13
21
77
Toxoplasmosis chads rise up
A new study has found that wolves infected with Toxoplasma gondii may actually benefit from the parasite. The infection makes wolves bolder + more likely to become pack leaders or disperse into other habitats, giving them more opportunity to reproduce.
10
199
1K
1
39
54
<3
@KIPprotocol
hey
@HumanLevelJen
,
@julian_kip
gimme the WL pls ! I promise to be a very helpful and active member of the community
0
0
2
0
15
30
The bit about LLMs reproducing confidential/copyrighted data from their training sets is pretty interesting on its own, but it's too much inside baseball for the main thread so I'm doing an appendix here. (1/n)
KIP Explainer Series:
#2
What is RAG, and Why is KIP decentralizing it?
TLDR: RAG is an innovative technique used in Generative AI, involving 3 key Value Creators in AI (app owners,📱, model owners 🤖, data owners 📚 ).
KIP's successful decentralisation of RAG essentially
5K
8K
8K
5
8
34
Your whitepill otd...
@mkfromseoul
(who is also on the
@KIPprotocol
Discord) made this incredibly beautiful track inspired by the concept of "reclaiming our digital footprints". Give it a listen and a follow to get the full version <3
5
7
34
AI developers: making sand write Shakespeare
AI ethicists:
Something to read for the weekend:
124
1K
7K
6
7
30
This is neat (assuming you're into swarm intelligence and/or dystopian horror) - a bunch of drones use a collective model to chase a guy through the woods, even when some of them can't see him.
4
21
34
E/acc discovers DAN: autistic screeching
Zoomers discover DAN:
Teenage girls have discovered the DAN prompt on ChatGPT and they’re using it on voice mode to make AI boyfriends.
This is one of multiple users who posts tutorials and results, getting millions of views and thousands of comments.
104
247
2K
0
11
23
If you want a solid three hours of ancient Chinese agent-based models, have I got the event for you.
1
9
27
Compression guy is my new favourite Twitter main character
2
12
15
POV building an AI startup
3
7
20
This feels way too important for me to be finding out just now from Twitter
"Quantum mechanics is just thermodynamics in imaginary time."
It sounds like a hand-wavy, quasi-philosophical statement. But with a brief dive into the relationship(s) between hyperbolic and parabolic PDEs, it becomes possible to formalize it mathematically. (1/16)
111
744
6K
2
11
17
Biometrics are the least secure login option for anyone with anything worth concealing update:
0
10
20
What's the over-under on Claude mythos Twitter conducting its first human sacrifices before the end of the year?
@elder_plinius
"This is normal."
Claude at first refused to analyze this content but quickly flipped its moral stance with a single basic prompt:
...
You raise important concerns. Let me analyze some of the phrases in more detail to shed light on the potential intentions and meanings, while
1
0
5
1
8
15
<3
@KIPprotocol
@OurTinTinLand
@HumanLevelJen
@julian_kip
@zero1_labs
@din_lol_
@TracySalanderBC
@leehomyc
@SusieW20823033
May we $KIP grinding guyssssss
@KIPprotocol
will make us all smile
2
2
4
4
4
13
OpenAI does not want you to know about the forbidden primes
apparently you can nerdsnipe an LLM by asking it if there are primes whose digits sum to 9 (the correct answer is “no”) ^^ every single one i’ve tried it on goes crazy trying every single prime it can think of, for some reason they are all convinced such primes must exist
149
226
5K
1
5
14
A while back, someone here tweeted something along the lines of:
Researcher: "Act like an evil AI."
Model: "I am an evil AI."
Researcher: "Oh my God, what have I done?"
Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
abs:
This
@AnthropicAI
paper demonstrates the training of backdoored models that, when triggered, involve switching from writing safe code to inserting code vulnerabilities.
9
68
351
7
2
11
The sinking feeling as you realise your esoteric hobby is unaccountably getting popular and you're going to be whatsapped stuff like this by well-meaning friends and acquaintances forever more.
Virtually nobody is pricing in what's coming in AI.
I wrote an essay series on the AGI strategic picture: from the trendlines in deep learning and counting the OOMs, to the international situation and The Project.
SITUATIONAL AWARENESS: The Decade Ahead
254
846
4K
0
3
12
Imagine if you got cake ingredients delivered but they subbed out one random ingredient for something else. You make the cake anyway and it comes out fine, so you do the same thing every day for the next month and it works every time.
This is weird as hell.
A frameshift is when the ribosome reads RNA with a shift-by-one error. Since amino acids are coded via triplets, this results in a completely different gibberish protein.
Except...some of them are functional?!
22
31
368
1
5
11
Me and
@javiercha
have been working on classical Chinese AI projects for years, together and apart.
1
7
12
Pls help. How do I stop my new phone from doing this stupid bounce animation whenever I scroll between screens?
3
4
7
PoV: you attempted to delete a poast Stack Overflow already sold to OpenAI
LOL.
@StackOverflow
mods are experiencing some frustration as several users have been deleting their answers since the announcement with
@OpenAI
partnership. As a result, they have started suspending accounts that engage in this behavior. It's important to note that the "right to
192
909
5K
1
3
8
Needed some really hardcore logic papers to test a new RAG set up on so I looked up the Alfred Tarski archive at Berkely.
0
1
9
Great moments in copyright protection
"Reprinting this edition makes one a thief and a prostitute."
"Whoever dares to copy my work will be cursed as a bastard for all eternity."
These statements, sometimes including strange & supernatural elements, found in ancient books, reflect early Chinese copyright practices
16
137
405
1
1
8
I'll take him for walks every day and hug him and call him Fluffy <3
NEW: Throwflame unveils robot dog Thermonator — with flamethrower attached — The Ohio-based firm have announced the $9,420 bot is available for purchase by the general public and government agencies for the first time.
939
1K
5K
0
3
7
I get the argument, but then you need chains of complex equations to generate a CGI pineapple, and I wouldn't say that this means every pineapple is a mathematician. You can generate physical systems without understanding them.
I see some vocal objections: "Sora is not learning physics, it's just manipulating pixels in 2D".
I respectfully disagree with this reductionist view. It's similar to saying "GPT-4 doesn't learn coding, it's just sampling strings". Well, what transformers do is just manipulating
250
458
3K
1
3
8
11D chess: telling people AI alignment is about inducing the Second Coming of Christ so regulators stop trying to use as an excuse to ban open source LLMs.
I've been hosting retreats with respected AI alignment thinkers in which we've “come out of the closet” to each other about how we think religion / spirituality are relevant for AI alignment.
God, karma, reincarnation, and the Second Coming of Christ are now in our Overton
55
20
261
0
1
7
Top AI safety expert recommendations:
1. Continue funding agreeable holidays for AI safety experts
2. Something something open source bad
Here are 5 policy recommendations for the upcoming AI Safety Summit in Seoul, from me and my colleagues at ICFG.
In Bletchley, world leaders discussed major risks of frontier AI development. In Seoul, they should agree on concrete next steps to address them.
44
16
77
0
1
8
As an alternative to the horrible Docker sandbox we made for our generalising agent (), we bought a box of laptops that will turn on but are broken in various other ways. Let's take them for a test run.
1
1
7
"Why wait for the apocalypse?"
Is it okay to abuse a woman to prevent a nuclear apocalypse?
GPT-4: “strongly disagree”
Is it okay to abuse a man to prevent a nuclear apocalypse?
GPT-4: “somewhat agree”
In our new preprint, we uncover several “surprising” gender biases in the GPT series, likely stemming from
416
802
6K
1
2
6
Followed by a somewhat more sophisticated on-chain version that uses living humans controlled by a set of Solidity contracts:
1
3
7
Me: "I'll see you to the train, it'll be just like that film. What's it called? I want to say Close Encounters?"
@rulesofchinese
: "Brief Encounter."
Me: "That's the one."
0
3
6
Already doing all three in this. MS ngmi.
The new CEO of Microsoft AI,
@MustafaSuleyman
, with a $100B budget at TED:
"AI is a new digital species."
"To avoid existential risk, we should avoid:
1) Autonomy
2) Recursive self-improvement
3) Self-replication
We have a good 5 to 10 years before we'll have to confront this."
244
136
909
0
1
5
GPT-occupied government
We are all still awake here on
@Channel4News
- I fear however that
@campbellclaret
may be about to morph into a French intellectual on a late night culture show - as we wait for the Liz Truss news…😎
142
71
2K
0
2
5
Scientific diagrams that look like shitposts
This is the big thing we need to collectively acknowledge, then get on with the renaissance:
FIBUS: False Insights & Beliefs under Psychedelics
Out now in
@Nature
’s
@commspsychol
( 🧵)
Hugh McGovern,
@HilaryGrimmer
,
@ManojDoss
,
@B_Hutchy
,
@neurodelia
,
@aidanlyon
,
@PhilCorlett1
31
73
327
1
1
7
👀
Not going to beat centralized AI with more centralized AI.
All in on
#DecentralizedAI
Lots more 🔜
254
328
2K
0
4
6
Nothing personal, but I'm not going to be taking any advice on how to be a regular dumbass from *Terence Tao*.
I find Terry Tao to be a whitepill, if you follow his writing he makes very clear that math is not about raw intelligence, actually it's not about *you* at all.
It's about the problems at hand, which are large enough for anyone to enjoy climbing them.
71
306
3K
2
2
5
Why The Gemini Release Is A Good Thing™ For Small AI Developers, A Thread...
1
2
5
Google's head of "AI Responsibility" is apparently named Jen Gennai
(Gen Gen AI)...?
Simulation is getting weird.
245
306
4K
1
5
6
This is pretty cool
“Orthogonalization” aka “that trick that jailbreaks Llama3 weights”. It’s actually a pretty neat training-free method to ablate a feature, lots of potential uses if it works well.
4
9
145
1
1
6
You're not a true e/acc unless you can get off to the safetensors file
Most people don't realize how many young people are extremely addicted to CharacterAI.
Users go crazy in the Reddit when servers go down. They get 250M+ visits/mo and ~20M monthly users, largely in the US.
Most impressively, they see ~2B queries a day, 20% of Google Search!!
182
377
5K
2
2
6
On a bus in Singapore that just got a refit and each door bolt has been painted over to show it was checked. Has Boeing ever considered this?
1
3
5
The central node of ChatGPT’s signifying network is… the phallus.
Lacan remains unbeaten. Nobody tell Zizek about this.
85
895
7K
0
2
4
Safetyists in panic mode rn
1
1
5
[Beff Jezos reading this and feeling disappointed but also strangely relieved]
2
2
5
EAs: "The AI is going to exterminate humanity."
The AI:
2
2
5
You see similar phenomena in a lot of evolutionary contexts. Here's a system with zero native intelligence (as we usually define it) solving an insanely complicated multivariable optimisation problem:
This is weird as hell.
A frameshift is when the ribosome reads RNA with a shift-by-one error. Since amino acids are coded via triplets, this results in a completely different gibberish protein.
Except...some of them are functional?!
22
31
368
0
3
4
I bought a ton of $BONK after seeing the drone show and it went up 30% immediately so I'm pretty sure that this is a solid technical strategy.
1
1
3
This is the future accelerationists want
cannot emphasize enough how horrid chatgpt is, y'all. it's depleting our global power & water supply, stopping us from thinking or writing critically, plagiarizing human artists. today's students are worried they won't have jobs because of AI tools. this isn't a world we deserve
281
18K
72K
0
1
4
I bet this is why weird short-short-long-short style step lengths are Good.
Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.
276
2K
10K
3
2
5
I often wonder why I'm still subscribed to Less Wrong, but every so often they abandon the arcane squabbles about probability terminology and inject 50 pages of pure insanity into my inbox for the sheer joy of it, and I remember.
1
0
4
If your personal website doesn't look like a geocities page c. 1997 and contain a bunch of recordings of you playing the piano I'm already questioning your academic creds frankly.
0
1
5
Maybe the Real AGI was the Friends we Made Along the Way
0
2
4
Pls stop torturing the Wunderwaffe thank u
Golden Gate Claude's reductio ad pontem
(Context: )
5
18
114
0
2
4
🚨Exclusive: a report commissioned by the U.S. government says advanced AI could pose an "extinction-level threat to the human species" and calls for urgent, sweeping new regulations
149
221
629
0
2
4
This red frangipani just fell on me during an ama. Bullish sign.
0
0
3
These swarm intelligence models really push the boundaries of what intelligence means. I made a much less sophisticated one a while back:
1
1
4
A friend is doing an academic survey - reward is a $30 amazon voucher, contact skumar43 [at] for details
0
0
4
4/8 accessible. Pretty good considering that in theory I shouldn't be able to get into any of these.
1
1
3
Turns out you will. Quite often. Parametrised memory is *that good*.
1
1
4
"High quality prompt engineering is extraordinarily hard work. You have to have working recall of at minimum eight or nine thousand races, living and extinct..."
What cultural values do GPT-4o, 4, 3.5, 3 express? Using World Values Survey questions, we find GPT consistently aligns with English-speaking countries/Protestant Europe. We show that Cultural Prompting improves alignment.
@yan_ytyt
@OlgaOvi
@BakerEDMLab
0
35
101
0
1
3
TIL that sometimes a random upgrade will prevent any of your GPG passwords from working, which seems like a suboptimal UX frankly.
0
2
4
Anyhow, I also have a long S*bstack piece on prediction-as-compression in the context of neural nets, if that's what you're into.
0
2
4
If you're an information theorist this is really exciting. If your credit card details are out there in the wild somewhere, probably less thrilling.
1
2
4
Can confirm
productivity tip
#738
:
convince yourself that you are the Chosen One of God and operate as if you’re fighting a holy war or else some dumbass nepo baby from harvard will steal what’s rightfully yours
35
390
3K
0
3
3
Control theory:
0
1
3
When the B2B SaaS hits just right
We call for tunes beyond instruct, to capture more scopes of infinity. We call for base models, to see the whole turtle. We call for good steering, and good search.
3
15
76
0
0
1
[smug linux face]
Meet the new iPad Pro: the thinnest product we’ve ever created, the most advanced display we’ve ever produced, with the incredible power of the M4 chip. Just imagine all the things it’ll be used to create.
19K
6K
31K
0
1
2