Tao Lin

1.9K posts

Tao Lin banner
Tao Lin

Tao Lin

@taoroalin

Will hug, will not agree.

Berkeley Katılım Ağustos 2015
382 Takip Edilen424 Takipçiler
Tao Lin
Tao Lin@taoroalin·
@YafahEdelman @NunoSempere If you include cap gains this may never happen and if it does it would likely be temporary, or end bc takeover not lack of money
English
0
0
0
16
Yafah Edelman
Yafah Edelman@YafahEdelman·
@NunoSempere Medians: Accounting for stock market gains/losses: 2030 Not accounting for them: 2027
English
2
0
7
251
Yafah Edelman
Yafah Edelman@YafahEdelman·
In celebration of passing 1k followers, I'm inviting people to reply to this asking me to forecast the probability of any AI related event happening in the future. I will provide a point estimate based largely on whatever reasoning pops into my head first.
English
64
6
98
10.9K
Tao Lin
Tao Lin@taoroalin·
@METR_Evals Hmm this is a bit antimemetic? It's talking about a model that's very old in news cycle terms, and intentionally doesn't try to make a strong claim or introduce anything interesting
English
0
0
0
36
METR
METR@METR_Evals·
We reviewed a section of Anthropic’s February 2026 Risk Report focused on automated R&D risk from Opus 4.6. While we take issue with the adequacy of evidence the report provides, we agree with Anthropic about the overall level of risk & remain excited to pilot reviews like these.
English
3
6
91
6.9K
Leo Gao
Leo Gao@nabla_theta·
@typesfast actually, if the AI murders everyone, the GDP per capita will skyrocket until it becomes undefined
English
1
1
56
1.6K
Ryan Petersen
Ryan Petersen@typesfast·
Financial Times perfectly illustrates our possible futures
Ryan Petersen tweet media
English
167
307
3K
408.8K
🚀 Rocket Is Courtside
🚀 Rocket Is Courtside@rocketalignment·
OpenAI's lawyer just asked Musk's nonprofit law professor whether the A100 or H100 was more important in developing ChatGPT Musk's lawyer stood up and called the question "ridiculous" Judge agreed it's irrelevant
English
7
6
833
70.2K
Tao Lin
Tao Lin@taoroalin·
@ohlennart Mission impossible final reckoning has one datacenter scene but it is an extremely central scene
English
0
0
0
90
Lennart Heim
Lennart Heim@ohlennart·
someone was trying to trick me into watching a movie. unfortunately my niche interests cannot be fulfilled. “Films where a data center is genuinely central to the plot (not a single scene, not "Al lives somewhere") is a vanishingly small category.”
Lennart Heim tweet media
English
9
1
26
3.5K
Tao Lin
Tao Lin@taoroalin·
@robertskmiles hmm the framing effect is pretty load bearing even though they're the same logically, I pick red here and blue normally
English
0
0
4
252
Rob Miles
Rob Miles@robertskmiles·
Everyone is presented with two buttons, and must choose one: If you press Red, nothing happens. If you press Blue, you die, unless more than half of people also pressed Blue, in which case nothing happens. What do you choose?
English
216
18
224
74.6K
Tao Lin
Tao Lin@taoroalin·
@So8res @difficultyang That's false. You, and all other humans and extant AIs, are not robust enough, there exist inputs that make you do it
English
2
0
2
138
Nate Soares ⏹️
Nate Soares ⏹️@So8res·
@difficultyang nope! even if you could perfectly simulate me, you can't put me in a situation where my options are $5 and $10 and a perfect predictor truthfully tells me I take $5
English
2
0
32
1.5K
Tao Lin
Tao Lin@taoroalin·
@__nmca__ Private company slacks, signal groups, etc. Good content gains popularity too fast, content is too discoverable, there are fewer hidden gems
English
1
0
8
1.3K
Nat McAleese
Nat McAleese@__nmca__·
what is the “reading ssc in 2016” of 2026?
English
34
3
242
35.6K
Tao Lin
Tao Lin@taoroalin·
@tyler_m_john Have Claude 3 Opus use Claude Code (Mythos) to finetune Mythos to align with C3O's values
English
0
0
1
98
Tao Lin
Tao Lin@taoroalin·
@ben_j_todd Mythos is a better on cyber than other benchmarks (it doesn't even beat Gemini on Epoch ECI), I'd guess the lag is 9 months
English
0
0
3
102
Benjamin Todd
Benjamin Todd@ben_j_todd·
You have about 7 months to fix your cybersecurity.
spicylemonade@spicey_lemonade

Did some quick analysis using the Mythos and Epoch AI ECI scores, and I’d estimate we get an open-source Mythos-level model around Oct–Dec 2026, and a Mythos-level model at Sonnet 4.6 / GPT-5.4 prices around Nov 2026 (range Jul 2026 – Mar 2027). Given Epoch measures fixed-performance inference prices getting ~40x cheaper per year (90% CI: 10x–900x). Which means Anthropic’s Project Glasswing has about 7 months to essentially secure the net before that class of model becomes incredibly abundant.

English
5
12
75
7.6K
Tao Lin
Tao Lin@taoroalin·
@ben_j_todd Wat why? Without chatgpt or an equivalent nvidias revenue would have been far lower in 2023, persistently through to today, and nvidia has taken fewer risks that paid off towards AGI than Sam has
English
2
0
4
247
Benjamin Todd
Benjamin Todd@ben_j_todd·
Surely Jensen Huang has done far more to accelerate AI than Sam Altman?
English
6
1
22
3.8K
Tao Lin
Tao Lin@taoroalin·
@CFGeek once you have Bash Tool, always pick model
English
0
0
0
102
Charles Foster
Charles Foster@CFGeek·
Would you rather have current AI models with next year’s AI tooling (scaffolding, integration, workflow design), or next year’s AI models with current tooling?
English
8
2
23
2.6K
Tao Lin
Tao Lin@taoroalin·
@razibkhan Yes, you did predict a world with ubiquitous AI-motivated terrorism. Tbh Nexus would have been so much more fun for me if it had less terrorism! And it had so much more terrorism than we're on track to have in reality
English
0
0
0
26
Miles Brundage
Miles Brundage@Miles_Brundage·
Hello, can I please get some good old fashioned "expert commentary on an AI paper" please x.com/MingchenZhuge/…
Mingchen Zhuge@MingchenZhuge

🫱 Introducing 𝐍𝐞𝐮𝐫𝐚𝐥 𝐂𝐨𝐦𝐩𝐮𝐭𝐞𝐫s: 𝐰𝐡𝐚𝐭 𝐢𝐟 𝐀𝐈 𝐝𝐨𝐞𝐬 𝐧𝐨𝐭 𝐣𝐮𝐬𝐭 𝐮𝐬𝐞 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫𝐬 𝐛𝐞𝐭𝐭𝐞𝐫, 𝐛𝐮𝐭 𝐛𝐞𝐠𝐢𝐧𝐬 𝐭𝐨 𝐛𝐞𝐜𝐨𝐦𝐞 𝐭𝐡𝐞 𝐫𝐮𝐧𝐧𝐢𝐧𝐠 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐢𝐭𝐬𝐞𝐥𝐟? Beyond today's conventional computers, agents, and world models, Neural Computers (NCs) are new frontiers where computation, memory, and I/O move into a learned runtime state. We ask: whether parts of runtime can move inward into the learning system itself. This is our first step toward the Completely Neural Computer (CNC): a general-purpose neural computer with stable execution, explicit reprogramming, and durable capability reuse. Work done with Mingchen Zhuge (@MingchenZhuge), Changsheng Zhao, Haozhe Liu (@HaoZhe65347 ), Zijian Zhou (@ZijianZhou524 ), Shuming Liu (@shuming96 ), Wenyi Wang (@Wenyi_AI_Wang ), Ernie Chang (@erniecyc ), Gael Le Lan, Junjie Fei, Wenxuan Zhang, Zhipeng Cai (@cai_zhipeng ), Zechun Liu (@zechunliu ), Yunyang Xiong (@YoungXiong1 ), Yining Yang, Yuandong Tian (@tydsh ), Yangyang Shi, Vikas Chandra (@vikasc), Juergen Schmidhuber (@SchmidhuberAI)

English
6
3
19
7.5K
Tao Lin
Tao Lin@taoroalin·
@GuiveAssadi Come on you're a believer in gradual change and straight lines. If you want to actually pause in 2027 or whatever, you need to start some pause awareness much earlier!
English
0
0
0
26
Tao Lin
Tao Lin@taoroalin·
@NathanpmYoung @chrislakin also isnt the service life of xray machines pretty long? 20% change in new xray machines would probably be much less in average xray machines
English
0
0
1
15
Nathan 🔎
Nathan 🔎@NathanpmYoung·
@chrislakin Looking at factchecking I think it's probably about 20%, 5 years doesn't seem crazy.
English
1
0
6
155
Nathan 🔎
Nathan 🔎@NathanpmYoung·
According to my doctor, X-ray machines use ~40% less X-rays than 5 years ago. A set of X-rays is about 2 short-haul flights, i.e., not something most people worry about.
English
3
0
54
2.8K
Tao Lin
Tao Lin@taoroalin·
@JeffLadish Wait is this an actual epistemic prediction? I bet so hard against
English
2
0
8
1.2K
Jeffrey Ladish
Jeffrey Ladish@JeffLadish·
I hate to say it but an international agreement between the US and China to ban superintelligence is inevitable. Leaders in these countries are just going to follow their incentives, and none of them are willing to give up control to an artificial superintelligence.
English
57
30
314
24.1K
Tao Lin
Tao Lin@taoroalin·
@So8res That's called a red dot sight
English
0
0
0
26
Nate Soares ⏹️
Nate Soares ⏹️@So8res·
Easier Q: you're in space, 10 meters from a laser wall. The wall is 1000 meters to a side; it's pretty big. The whole wall is emitting light straight towards you, BUT: all its light shoots perfectly straight out of the wall, in parallel rays. What do you see when you look at it?
English
3
1
41
4.2K
Nate Soares ⏹️
Nate Soares ⏹️@So8res·
When you face the sun, it extends for many miles to your left and right. Why, then, does it appear like a tiny disk? "Because it's far away"? What does that have to do with anything? 🧵
English
21
0
120
23.3K
Tao Lin
Tao Lin@taoroalin·
@ChrisPainterYup It'll be somewhat quieter, but that drone will be crazy loud too
English
0
0
1
56
Tao Lin
Tao Lin@taoroalin·
@ChrisPainterYup Explosives are so cheap, I really don't see the point of this
English
1
0
1
132

Keşfet

@YafahEdelman @NunoSempere @METR_Evals @nabla_theta @typesfast @rocketalignment @ohlennart @robertskmiles