Clifton King
3.5K posts
Clifton King
@cliftonk
dad x3. investor. ex-dialectic, sixty capital, @bridgewater. saas entrepreneur. co-founder @atxdao
Austin, TX Katılım Nisan 2008
946 Takip Edilen997 Takipçiler
@cdriclion @nic_carter my main contention is that the harnesses we're using are profoundly primitive.
English
It's not - any agent using "harness (context management/compaction, subagents, forking, evals, etc)." that would perform long duration AGI without any human intervention, and be good at pursuing the goals we gave it (or it chose) over months and months without collapsing would be an AGI to me.
But I have not seen anything like that in action (although I'm sure it might happen very soon)
English
The “it’s not AGI because machine intelligence is jagged” is dumb cope.
It’s obviously AGI. If you had a friend who had a 130 IQ, could write production code flawlessly, could write academic papers of a high research caliber, pass any exam in any field with flying colors, create a sophisticate LBO model, draw technical diagrams perfectly, compose poetry in any language, and could find solutions to significant unsolved mathematical problems, you would call that person a world historical genius. Certainly, no single human has ever had intelligence that “general” before.
Now you think it’s “not AGI” because it sometimes slips up and makes mistakes - so does any human that you would consider “extraordinarily intelligent.”
The professor might forget a colleagues name that he has known for a decade. He is still considered intelligent. The math genius might be a little autistic and shy, unable to maintain polite conversation. Still intelligent. You might stare at the fridge for 30 seconds unable to find the butter, despite 5 million years of evolution perfecting your visual intelligence.
We give intelligent humans a pass when they have jagged intelligence. So why the double standard?
The qualities people list as “necessary for AGI” are important traits to have, but no longer pertain to intelligence. People will say things like “true AGI requires agency, long term goal setting, embodiment, self-direct action”.
But none of those things are intelligence. Those are “things that humans have that AI lacks”. Raw intelligence, AI has it in spades. That other stuff - important yet, but broader than and different from intelligence.
The unwillingness of people to acknowledge that AGI obviously exists and has existed for a while is due to a kind of anthropic chauvinism - a psychological need to believe that humans are superior in every respect, that we possess soft skills that no machine could replicate.
Yes humans are different from machines, but if we are limiting the discussion solely to general intelligence, AI has it already. That battle is over.
If you want to reframe the discussion to matters of human dignity and personhood, fine, but that’s not an AGI question. That’s something else. Just take the loss on AGI already. It’s over.
English
@cdriclion @nic_carter if your criteria is that a limited-context, high dimensionality next token predictor should be able to perform long duration AGI then maybe there's no point in arguing, because in that case i would also agree with you.
English
@cullenroche @TheStalwart if the benefits begin to appear as fast as we think they will, the reflexive demand for energy and raw materials will be insatiable. its possible services disinflation counters it for americans, notsomuch elsewhere.
English
Let me ask my AI. Ha.
I feel like it's all happening faster than I expected. I don't even think the inflation effect near term is that big. This bump in inflation is mostly the dumb war effect. Back that out and inflation is pretty benign.
Did you see this piece at Liberty Street? It's good.
AI's Macroeconomic Challenges and Promises - Liberty Street Economics libertystreeteconomics.newyorkfed.org/2026/05/ais-ma…
English
This isn't my view at all. We're in a two phase AI build out. Phase 1 looks inflationary because the build out causes harsh bottlenecks as we scale super fast and wealth effects impact market prices.
Phase 2 is the legacy of Phase 1s build out. It will result in a supply side boom when robots are making things 24/7 and unit labor costs collapse. This is hugely disinflationary in the long run.
We're still in Phase 1, but Phase 2 is coming. It's not if, it's when.
Joe Weisenthal@TheStalwart
"AI is inflationary and it's only a matter of time before Warsh is forced to hike rates" is becoming a completely conventional wisdom view among money managers. Here's part of @ozanktarman's summary of his latest macro dinner
English
@cdriclion @nic_carter what you are saying the models are lacking can entirely be solved via harness (context management/compaction, subagents, forking, evals, etc). they can even modify their own harness on the fly (see Pi).
"AGI" is a moving goalpost. we've had it for some time now.
English
Well, without that, it remains fundamentally less "capable" than humans, because not autonomous.
AWS seems to include autonomy in it's definition of AGI at least (first result I get from typing AGI on google) :
"AGI is a theoretical pursuit to develop AI systems that possess autonomous self-control, a reasonable degree of self-understanding, and the ability to learn new skills."
aws.amazon.com/what-is/artifi…
Of course you can argue that "they are autonomous to some extent", but the scale of this autonomy is still very unsatisfaying.
English
Agentic workloads are comically memory-inefficient. My personal info-aggregation agent is burning ~7GB of RAM just to process text for one person.
Clifton King@cliftonk
@brandonjcarl @fkronawitter1 here's my openclaw which i use primarily for information aggregation in my trading. its using 7gb of ram for *processing text* for a single person.
English
@brandonjcarl @fkronawitter1 here's my openclaw which i use primarily for information aggregation in my trading. its using 7gb of ram for *processing text* for a single person.
English
when you run a vibe coded application, it will spend about 95% of its time in I/O (network calls, reading/writing from disk, etc). the way every variable and datastructure are laid out in memory is *absurdly* inefficient. this is true for every script the system is creating/running (and even the harnesses themselves).
persistent applications created in these languages are even worse. the massive volume of code being generated (which i expect to 100x from here in the next few years) will use even more memory. open system monitor and see how many resources something like TradingView (a webkit app) uses vs a native app.
English
Memory is too expensive. Human ingenuity will find many ways to use less of it more efficiently
English
@brandonjcarl @fkronawitter1 bear case is that buyers form a consortium to make their own. regardless of advancements in efficiency. demand for memory will only go up. agentic use cases are typically on extremely memory-inefficient code (eg python, typescript) that need very little CPU and tons of RAM.
English
@fkronawitter1 I agree Florian. Good insights in this video.
This gets solved through Hybrid Linear Attention or similar. We’ve run into problems like this in the past (sorting, linear programming) and you always find a better algorithm.
Brandon Carl@brandonjcarl
Great insights from @JonathanRoss321 at Sohn. Investors look at "What is the bottleneck?" That's not how this industry works. Every time a bottleneck gets big enough, people solve it. youtube.com/watch?v=OBAXUd…
English
7x is growth is very bearish. Indicates essentially no one is using google routed models on advanced harnesses. Labs probably at 100-500x YoY. (Google still hosts the models, they just aren’t being routed to thru GCP billing).
sunny madra@sundeep
English
@RagingVentures We will see demand destruction everywhere except GPU/ASICs even if that’s the case for inference. Agentic workloads need tons of memory (which is scarce while CPUs are not).
English
Run, don't walk, to listen to this interview if you're investing in memory $MU $SNDK
NVDA's Ross argues that if memory prices stay high, engineers will devote a large portion of their time to solving the problem by using less memory
Eugene Ng@EugeneNg
Interview: youtube.com/watch?v=OBAXUd…
English
Clifton King retweetledi
Frog put the shares in an SPV. “There,” he said. “Now we can transfer these shares freely”
“But Anthropic can still exercise its transfer restrictions” said Toad.
“That is true,” said Frog.
_gabrielShapir0@lex_node
I am surprised more people are not paying attention to this update from Anthropic on its stock policy. This seems like a potential bombshell. There is an active secondary market purportedly in Anthropic stock or derivatives including on fairly reputable (or at least well-known) platforms like Forge. Anthropic is calling them out *specifically*, by name, and essentially *saying* 100% of these are illegal. Some may be frauds (people selling Anthropic stock or interests in Anthropic stock that they don't truly own), but more likely many are legit attempts at transferring Anthropic equity (directly, as SPV shares, or as some type of 'beneficial interest' or future, etc.) Anthropic appears to be saying it will treat all these transfers as void. I don't have access to their terms, but it's very interesting to think what this could mean. Do the 'first purported sellers' in the chain potentially have an opportunity to do a double-dip? Does the first seller and all downstream buyers get the entire entitlement nuked? Anthropic is threatening that--are they just bluffing? If they're not bluffing, what litigation is likely to ensue? This can get into really esoteric areas of corporate law that depend on exactly how the transfer restrictions are drafted as well as the language around how violations of transfer restrictions are treated--for example, if they are merely voidABLE then downstream buyers can assert various equitable claims/defenses, but if they are VOID ab initio then in some jurisdictions that forecloses equitable defenses.
English
@sundeep Another good indicator was how the goal posts kept moving as it became clear the models were already near-AGI capable and while the harnesses were lacking.
I still see contemporary papers on task coherency not even mention the harnesses…
English
@fkronawitter1 That said I have no view on where memory stocks end up. Could very well be near the top. 🤷
English
@fkronawitter1 I think we’ll see memory cause demand destruction in CPUs and consumer devices (CPUs and consumer devices ex Apple have little to no differentiation) long before we get more supply online. As crazy as this sounds I think more likely than not dram doubles again by eoy.
English
@RHouseResearch It’s bad for aggregate financial market liquidity. It’s entirely conceivable that hyperscalers grow enough to make them cheap in today’s dollars, but if liquidity deteriorates multiples will come down. Another inflationary episode also seems increasingly more likely.
English
"You've already going to be diminishing the buybacks because of all of the commitment to capex from the hyperscalers. They're already going to be eating into their cash flow."
This narrative that hyperscaler capex investment will be a major market headwind because it reduces cash available for buybacks falls flat on its face when you look at just a few facts.
First, the hyperscalers themselves. Google has fully halted buybacks and is up 22% YTD and at all time highs.. Meanwhile Microsoft's buybacks are unchanged YoY yet the stock is still down 12% YTD. The return on capex investment is the driver, as Google is rewarded for capex (GCP +63% YoY rev. growth, AI mode boosting search, etc.), while Microsoft hasn't invested enough to enhance their existing offerings with AI (e.g. M365 Copilot at <5% penetration).
Second, hyperscaler capex = free cash flow for NVDA (and others)... which is used to buyback stock. At the peak of hyperscaler buybacks in 2022, Amazon, Microsoft, Google, and Meta bought back a combined $126B of shares. In the last twelve months, NVDA alone has bought back $40B of stock, and those buybacks should ramp materially this year with FCF estimates at $150B+ (and mgmt. intending to return 50% to shareholders).. Broadcom bought back more stock last quarter ($7.9B) than any year in its history.. Etc. Etc.. So while the hyperscalers themselves are buying back less stock, the businesses on the receiving end of their capex (semis, memory, etc.) are picking up the slack.
$AMZN $MSFT $GOOG $META $NVDA $AVGO
Patrick OShaughnessy@patrick_oshag
Paul Tudor Jones: "2000 was the easiest bear market I've ever seen in my whole life. It's got so many similarities to right now."
English
@buccocapital Income on X and kids on Y is a perfect midwit chart. It’s quite literally a semicircle.
English
Interesting that if you have three kids you’re probably either very rich or very poor
English