@Amank1412 @madhujith12 What is clearing his client queue doing though? All these vague posts
English
doffy
1.9K posts
@ghettokenn @ViktorBunin Or because they over hired and are now cutting to improve margin
English
@ViktorBunin Previous AI winters had impressive demos and no revenue. This wave has companies cutting headcount because the tools actually work. Wrong cycle to pattern-match against.
English
it is so fucking obvious we're not going to have AGI in the next few years. it's all a big psyop to make AI builders feel important and to justify the extravagant spend and valuations. everyone should calm down.
English
@jack @BarnieTaintpipe The bottleneck was never writing code and you know this
English
Some insider corroboration of my industry perspective from someone at Block who operated a more nimble team.
x.com/BarnieTaintpip…
Jason Karsh@jasonkarsh
English
@benjamincowen What happened to all the educational crypto influencers? The ones that taught the basics? I guess the interest just isn’t there
English
90% of crypto influencers now need to now find a new business model that does not just involve them pretending to like a project that they were paid to promote allowing them to dump their "allocations" on the people that trusted them
Nikita Bier@nikitabier
Today we're announcing Paid Partnership labels on posts. X's core value is providing on authentic pulse on humanity. While we want to encourage people to build their businesses on X, undisclosed promotions hurt the integrity of the product and lead people to distrust the content they read on X. This new feature will allow you to comply with regulations, but more importantly: it enables you to be transparent with your followers.
English
🚨#BREAKING: According to new reports documents tied to the Epstein files reveals that the FBI’s New York office was hacked in 2023 on the night of the Super Bowl, resulting in the reported deletion of approximately 100 terabytes of evidence data.
English
@froggy_dl You also don’t need a bunch of gimmicky stuff to give movement a high skill ceiling. Look at CS and the movement disparity between even pro players. I’m not talking about fancy ladder jumps or kz either.
English
@benjamincowen Unfortunately people only see the second one. The first one means CBDC’s. That’s the view from normies.
English
Being "pro-crypto" and "pro-how-can-I-extract-as-much-value-out-of-crypto" are two very different things.
English
17k tok/s is a shiny benchmark, but "hard-wiring" weights into silicon is a fundamental bet against the speed of ai research.
we’re moving from static inference to dynamic agentic loops. in sglang, we need the flexibility to optimize kernels and manage memory on the fly as models evolve. baking llama 3.1 into a chip in 2026 is like building a high-speed railway that only leads to a ghost town—by the time it's finished, everyone has moved on to the next architecture.
the real moat isn't raw speed on a frozen model; it's the reprogrammable efficiency of the serving stack. i’d rather have 5 lines of optimized, flexible code than a billion-dollar brick that can't change its mind.
Wildminder@wildmindai
17,000 tokens per second!! Read that again! LLM is hard-wired directly into silicon. no HBM, no liquid cooling, just raw specialized hardware. 10x faster and 20x cheaper than a B200. the "waiting for the LLM to think" era is dead. Code generates at the speed of human thought. Transition from brute-force GPU clusters to actual AI appliances. taalas.com/the-path-to-ub…
English
@InboxZero @mikubeam @chribjel So if they trained their own model then what copying was done exactly?
English
"distillation attacks"
so they payed for usage, and used the output?
basically what im doing
Anthropic@AnthropicAI
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
@ratechange1 @atmoio It’s easy to over extrapolate technologies we don’t understand
English
@atmoio everything happens on the margin and on the margin, white collar is fucked and will only increasingly be more fucked...but sure everyone you know will be fine
English
Why you won't lose your job to AI
Citrini@citrini
JUNE 2028. The S&P is down 38% from its highs. Unemployment just printed 10.2%. Private credit is unraveling. Prime mortgages are cracking. AI didn’t disappoint. It exceeded every expectation. What happened? citriniresearch.com/p/2028gic
English
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax.
These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
@JamshidHormuz @AbHomineDeus The world isn’t a bunch of tokens. It’s a bunch of states.
English
Bayesian models also predict future states probabilistically.
A transformer modeling P(next token | context) is doing the same thing.
Different architecture doesn't change the objective. it's still prediction.
And improving prediction requires accurate models of how the world works.
English
"AI isn't thinking! It's just pattern matching and predicted the most likely outcome." My Brother in Christ what do you think thinking is?
English
@AbHomineDeus So the cure to cancer will be found by predicting the most likely next token? Definitely not. Otherwise we would’ve found it by now.
English
@JamshidHormuz @AbHomineDeus It’s definitely not predicting the future. Look up Bayesian models. Directed Acyclic graphs can predict outcomes. ChatGPT is forward feeding and it’s more like walking on a footpath through the nature reserve. Actual thinking involves walking off the beaten path.
English
@AbHomineDeus Next token prediction = Predicting the future
The craziest part is that they are actually saying "AI isn't thinking, it just figured out how to predict the future" like this isn't incredibly impressive
English
Ya it’s tough for me to hop on the AI replacement train. As someone who’s coded generative forward feeding neural networks since 2023 it’s pretty obvious the limitations. Worse customer support, can’t work in large codebases. Individuals can be more empowered but that’s not job replacement.
English
Interesting read, certainly something to consider. Even somewhat aligns with the business cycle views I expressed here:
youtube.com/watch?v=2ehTz4…
I think there is a reason to believe the current business cycle will end within the next 2-3 years from a liquidity point of view.
YouTube
Citrini@citrini
JUNE 2028. The S&P is down 38% from its highs. Unemployment just printed 10.2%. Private credit is unraveling. Prime mortgages are cracking. AI didn’t disappoint. It exceeded every expectation. What happened? citriniresearch.com/p/2028gic
English
Anyone remember back when the replies were crypto scam bots?
Now they are just AI bots
English
@mattshumer_ @Curlh1 LLM performance is compute bound O(N^2). We’ve known this since 2017
English
@Curlh1 Probably engagement farming... well, it worked :)
Gets him likes but will hurt people who will believe him and ignore the (obvious) progress.
English
This is completely wrong.
Try the same difficult prompt on an LLM from a year ago and a frontier LLM from today. It's not even close.
Burying your head in the sand is comfortable, but not helpful long-term.
Make up your mind after trying the best AI systems available today.
Universal Public Defender@DefenderOfBasic
it's WILD that no one has noticed that LLM progress has stalled. All the signs are there
English