Filter
Exclude
Time range
-
Near
It genuinely feels great being left behind, making stuff through my own ability and being able to analyze and think about things on my own without offloading my thoughts to a machine that calculates the probability of word pairings.
3
Replying to @TimGarrisonMN
If he wasn’t from Minnesota 90 percent of fans wouldn’t want him. If you have a lineup of Jaden Rudy and Suggs you are setting ant up for failure. Hard hard pass unless he’s the only guy you can get by offloading Julius
11
The Moat That Eats Itself Satya Nadella's "A frontier without an ecosystem is not stable" arrives as economic analysis and behaves as strategic positioning. The surface argument is clean. Companies build two forms of capital, human and token; the durable edge comes from a proprietary learning loop on top of switchable models; concentration of value in a few models invites backlash, so the future belongs to a distributed frontier ecosystem. Read it twice and the seams show. The piece holds one real insight, then asserts a conclusion its own premises will not carry. The test is plain: does the reasoning earn the title, or does the title arrive by assertion? Start with what holds. The distinction between offloading a task and offloading your learning is genuine and well put: you can hand off the work, but the judgment that decides which work matters does not leave with it. The complementarity claim, that human capital grows more valuable as token capital grows rather than less, is a defensible reading of how firm-level capability builds, grounded in the economics of complementary assets. So the foundation is sound. If the floor holds and the roof still leaks, the fault is in the structure between them. That structure breaks at its load-bearing joint. Nadella opens by warning that frontier models can "continuously absorb the expertise of humans and organizations and commoditize it." He closes by promising a learning loop that yields advantage "hard to replicate, regardless of any new individual model capability." A careful defender splits the apparent clash into two bodies of knowledge: the general expertise that commoditizes, and the firm-specific loop that stays protected because it trains inside your own reinforcement environment, on your own traces, which never feed the shared model. Grant that in full. The loop still has to run through a substrate model at inference, and if that model belongs to a third party, the "company veteran" speaks through the provider's infrastructure every time it runs. Whether those traces are absorbed is set by the provider's data policy, the dependency that "sovereignty," defined as the freedom to switch models, leaves out. So the burden sits with Nadella. He posits continuous commoditization as the background condition; he owes the account of why the firm's loop escapes it, and never gives it. An unmet burden is enough: the durability claim rests on a separation the essay assumes rather than earns. Grant the moat anyway, and a flat contradiction surfaces. Here the text does not merely owe a proof; it asserts two things the engine cannot reconcile. The compounding loop Nadella sells as broadly distributive is, by his own description, a concentrating engine. Early builders gain an advantage that is hard to replicate; every improved workflow generates better training signal; the "hill climbing machine" compounds where most assets do not. Those are the mechanics of winner-take-most. A defender will place the promised distribution on another axis: value spread across firms and industries rather than captured by a few models, firm against provider, not firm against firm. The engine concentrates on that axis too: between firms, where whoever starts ahead compounds fastest, and back at the provider, since the loop only keeps climbing while bolted to one orchestration stack. By the rule this essay applies to Nadella, that bolt has to be named: the loop's machinery is tuned to one tooling layer and in practice does not port across stacks the way a model ports, so the portability he sells at the model layer goes missing where the value pools. The provider axis holds on the most generous reading of distribution, which is what the value layer below makes concrete. So the "stable equilibrium" he wants is undercut by the engine he prescribes to reach it. Stable does quiet work here, sliding between persistence, a configuration that holds because no actor gains by deviating, and preference, a configuration we ought to want. Compounding favors the first mover on either axis, the opposite of the distribution the conclusion claims. Ask the strategist's question, where does value accrue in the stack, and the framing explains itself. The interest is not the absence of a frontier model, since Microsoft builds those too, through OpenAI and its own MAI line. It is a competitive, interchangeable model layer beneath the orchestration stack the company sells, Microsoft as one of several swappable suppliers below and the sole landlord above: the platform, the compute, the evaluation and reinforcement tooling, the orchestration around the model. "Sovereignty," defined narrowly as the freedom to switch models, is the tell. It counts the one dependency the architecture removes and ignores the ones it leaves in place. You own the loop; you rent everything the loop runs on. None of this refutes the logic. Both findings were settled from Nadella's own sentences before motive entered, and they would hold if a disinterested author signed the same words. A claim is not false because the speaker profits from it; the motive only tells the reader how to weight the word should. The essay presents as inevitable and desirable the market structure most favorable to a platform incumbent. Which returns us to the title. By the essay's own mechanics, the ecosystem it describes is no more stable than the concentration it warns against, because the loop that encodes a firm's knowledge compounds fastest for whoever starts ahead. The analogy to outsourcing gestures at a real fear, and it cuts the other way once you name the mechanism: platform-mediated ecosystems pool value at the platform layer, the way outsourcing pooled it in the metropole, so the cure carries the disease it treats. The piece works better as a map of the terrain Microsoft would prefer than as a theorem about where the terrain must go. The question it never answers is the one that matters most: what keeps your loop yours, once it speaks through a model you do not own? ousadia criativa. precisão estratégica. – por kim.
24
Replying to @UnTalNixon_exe
Buen apunte. El ancho de banda sí es el cuello de botella real más allá de la capacidad de memoria.El Ryzen AI Max 395 tiene 256 GB/s teóricos (LPDDR5X-8000 en bus de 256 bits), no ~500 GB/s. El Infinity Cache ayuda un poco en ciertos patrones, pero sigue estando bastante por debajo de una GPU dedicada de gama alta o de los Apple Silicon Max/Ultra. Dicho esto, su gran ventaja sigue siendo la memoria unificada de 128 GB accesible directamente por la iGPU. Eso permite correr modelos que simplemente no entran en una sola RTX 4090/5090 sin offloading lento. Para modelos de 8B a ~32-40B (especialmente quantizados) a buena velocidad y con contexto largo o RAG, es una opción muy interesante y accesible.
42
Replying to @AdelDeveloperX
Automating a week's work while sleeping actually sounds wild. Been wiring vision models to delta arms lately, and offloading calibration loops to AI saves serious hours. One focused hour really does compound.
1
5
tic toc tic retweeted
@TicTocTick Your newsletter on $SPCX about mobilizing retail and offloading debt was another great read. I look forward to these, the only newsletter I read twice.
4
1
17
2,478
You have probably read about cognitive offloading. But have you given any thought about the thoughtless physical offloading happening around us? A few thoughts on my substack. Link in the comments.
1
2
koko retweeted
there is something spiritually wrong with living life on autopilot and never thinking for yourself. offloading your opinions, taste, thoughts, feelings and actions onto others. giving in to things without thinking them through to satisfy society norms and please others
3
6
139
4,469
Kat retweeted
Replying to @BenjaminNorton
It is worse than the crypto, $BTC $DJT pump & dump. They are offloading trillions in toxic AI stocks, debt and systemic risk onto retirement accounts. This is so bad.
Replying to @Kalshi
$ORCL $GOOGL $SMCI all raising capital to survive. Trump & billionaire buddies are herding retail from one bubble to the next for exit liquidity. They needed a fake compute shortage narrative prior to toxic #AI IPOs 👉They created a crypto $BTC frenzy for WH donors like Peter Thiel and Musk to exit. 👉 Trump used a fake “#Bitcoin treasury” pump to dupe millions of gullible idiots. 👉 Now they created an AI frenzy by creating a fake “compute shortage” to create an AI frenzy. $SPY $SOXX $SMH $INTC $NVDA will repeat the $BTC Pump and dump pattern ( & $ARM $AMD $AVGO $MU $SNDK) They need an Ai bubble to offload toxic assets onto retirement accounts at exorbitant multiples. They are offloading systemic risk into $SPY $QQQ $VOO $DIA indices. OpenAI is pre-BK, Anthropic worth 1/10th IPO price projections. If you are over 40yrs of age needs to GTFO & text everyone you know to do the same.
2
4
25
1,165
Replying to @Beezie
@OldTownJB in case you still have those slabs you’re offloading
51
Replying to @YashRMFC
Offloading mayb players will make it a 10
84
Offloading
1
1
Good luck in offloading this contract in the years to come. Alaba & Mariano situation vibes.
4
Offloading and adherence through technological advancements: Modern approaches for better foot care in diabetes #ActAgainstAmputation #ALPSlimb #NIDDK #SmartBoot diabeticfootonline.com/2024/…
23
Replying to @Agenda_Fc_
It's not pain brother It's me offloading a big burden
1
8
Fire break out🚨🚨🔥🔥🔥🔥 in umuahia today , nkwoegwu ohuhu Aka god filling station , according to the eye witnesses the fuel truck was offloading while it caught fire 🔥 🔥 🔥 🔥 🔥 🔥 🔥 🔥 🔥
7
Our recent business includes us taking Hazard, Courtois, Rudiger in their primes. And us offloading Kovacic to you
1
30
Replying to @veerszaara
im so close to offloading
12