与「Two_Three」相关的搜索结果

【モーニング娘。】モーニング娘。’22 × DA PUMP KENZO「One・Two・Three」夢のコラボ【OH舞DA PUMPエボリューション】 @YouTubeより やばばばばばばぁぁぁい！！！！！！

I don't think people understand just how bad it will be if an American open source champion doesn't emerge soon and the big labs succeed in creating modern East Indian companies and ban open models on moronic national security grounds. "If a credible Western open frontier player does not emerge, the consequences cascade quickly. This is the inverse of the early Internet wave. In the 2000s and 2010s, Western companies — Google, Facebook, Amazon, Apple, Microsoft — dominated globally while China carved out its own walled garden. "The AI version flips that dynamic on its head. "Without a credible Western open frontier player, the only open models capable of running entire economies are made in China. If U.S. policy further restricts Chinese open-weight access on national-security grounds, the U.S. ends up with two or three closed Cathedrals serving the U.S. market — and the rest of the world picks the AI stack that is free, capable, self-hostable, and not embargoed. "Europe, Africa, Southeast Asia, Latin America, India, the Middle East. "Roughly six billion people. "Chinese open models become the global default by 2030, and the United States ends up technologically isolated from the majority of the world’s AI users. "We would have done it to ourselves."

Funny enough, this is already the third post in my Waza skill design series. Today’s one is about /think, the skill I use for solution design before writing code. I have two settings in Claude Code that I find especially useful. The first is /model opusplan, which means planning is done with Opus while execution runs on the regular Sonnet model. That helps me save Max usage for the places where it matters more. The second is that I usually run Claude Code with alias c="claude --dangerously-skip-permissions". I would not recommend that to less technical users. I use it because I know what it is doing, and mostly because I am lazy. Back to /think. How do you get the strongest model to produce better technical plans? It starts with the model itself. Models tend to avoid taking a position. I prefer engineers who can give a clear recommendation. So the first thing I do is require the model to have a point of view. It must state its recommendation, explain what evidence could overturn it, and avoid empty lines like “There are many ways to think about this.” Giving two or three options is fine, but it has to make a clear recommendation, and it must always include a minimal option. But a plan is not done just because it sounds good. The second step is to make it argue against itself. Under what conditions would this plan fail? If those problems can be fixed, the fixes should be folded back into the plan and the revised version presented again. If the plan breaks under certain conditions, it has to say exactly where and why it fails. That way, by the time the plan reaches you, the tradeoffs are already visible. I also go fairly deep on validating the premises before planning starts. First, it checks whether it is even looking at the right part of the codebase. I have seen models produce plans against the wrong path. Then it looks for older technical design docs to avoid reinventing work that already exists. After that, it searches GitHub to see whether similar problems have already been solved elsewhere. Only after those three steps does it start proposing solutions. That helps prevent the entire plan from being built on a bad assumption. There is also complexity grading. If the work touches more than eight files or introduces a new service, the plan must explicitly call out the scale. If data flows across more than three components, it has to draw an ASCII diagram and look for cycles. API keys and third-party dependencies also have to be listed during the planning phase, so you do not waste time or end up with a plan that depends on shaky assumptions. There is one more hard rule. The plan cannot contain things like TBD, TODO, “we can decide this later,” or vague phrases like “similar to step N.” That goes back to model behavior again. Once you leave that kind of escape hatch, execution tends to drift, skip work, or fill in the blanks poorly. I try not to leave the model any room to wiggle out of precision. The output format is also strict: what we are doing, what we are not doing, which option was chosen and why, three to five decision factors, and a clear list of unknowns. /think does not write code. Execution only starts after the user approves the plan. When I built this skill, I was really trying to capture how strong technical experts approach solution design: investigate first, form a clear recommendation, make decisions decisively, leave no loose ends, and improve the plan immediately when something invalidates it. If you have better ideas for planning and solution design, feel free to contribute to Waza.

We are one step closer to having AI generate code better than humans! There's a new open-source, state-of-the-art code generation tool. It's a new approach that improves the performance of Large Language Models generating code. The paper's authors call the process "AlphaCodium" and tested it on the CodeContests dataset, which contains around 10,000 competitive programming problems. The results put AlphaCodium as the best approach to generate code we've seen. It beats DeepMind's AlphaCode and their new AlphaCode2 without needing to fine-tune a model! I'm linking to the paper, the GitHub repository, and a blog post below, but let me give you a 10-second summary of how the process works: Instead of using a single prompt to solve problems, AlphaCodium relies on an iterative process that repeatedly runs and fixes the generated code using the testing data. 1. The first step is to have the model reason about the problem. They describe it using bullet points and focus on the goal, inputs, outputs, rules, constraints, and any other relevant details. 2. Then, they make the model reason about the public tests and come up with an explanation of why the input leads to that particular output. 3. The model generates two to three potential solutions in text and ranks them in terms of correctness, simplicity, and robustness. 4. Then, it generates more diverse tests for the problem, covering cases not part of the original public tests. 5. Iteratively, pick a solution, generate the code, and run it on a few test cases. If the tests fail, improve the code and repeat the process until the code passes every test. There's a lot more information in the paper and the blog post. Here are the links: • Paper: • Blog: • Code: I attached an image comparing AlphaCodium with direct prompting using different models. 2024 has barely started, and we are making a ton of progress!

I asked Claude to apply a capital cycle analysis to $MU. Here's what it came up with: Net reading: 11 of 14 capital cycle signals are bearish or strongly bearish. The framework reads this as late-cycle, not early/mid-cycle. The two unambiguously bullish signals (equipment lead times, industry concentration) are eroding rather than strengthening. Insights Yielded by Capital Cycle Analysis: 1) "Structural change" rhetoric is itself diagnostic. The capital cycle framework treats coordinated industry-wide CEO claims of regime change as evidence of late-cycle euphoria. The same language was deployed by the same CEOs (Mehrotra at Micron specifically) in 2017–2018 and was wrong. Bayesian base rates argue against accepting the current claims at face value. The previous analysis under-weighted this base-rate evidence. 2) Look at total capital flowing into the supply curve, not just incumbent capex. The structural-change analysis focused on Big Three capex. The capital cycle lens forces aggregation of all capital flowing into memory output: a) Incumbent capex: ~$104B in 2026 across DRAM + NAND; b) CXMT IPO proceeds: ~$4.2B (with state-aligned co-financing many multiples larger); c) YMTC capacity additions (privately financed) d) Substitute technology capital (Cerebras, photonic startups, CXL controller designers) — billions of dollars of equity raised to reduce HBM intensity per dollar of AI compute deployed. When aggregated, total effective supply-side capital formation in 2026 is materially higher than the Big Three capex alone suggests. The supply response is being underestimated. 3) The customer base is doing exactly what late-cycle customers do. Hyperscalers locking in 3–5 year LTAs, pre-ordering 2027 NAND, building strategic inventory — these are not signs of confident long-cycle visibility, they are signs of late-cycle scarcity panic. Historically (DRAM 2017–2018, oil 2008, shipping 2007), customer pre-buying at peak prices is followed by sharp inventory destocking when prices roll over. The structural-change narrative frames LTA penetration as a benefit; the capital cycle frames it as a peak signal. 4) Multiple expansion + earnings expansion = asymmetric downside. The previous analysis flagged the 15x NTM P/E multiple as aggressive (referring to UBS PT raise). The capital cycle framework sharpens this: when both earnings and multiple are at peak, the compound drawdown when either reverts is severe. Memory historically goes from 60% gross margin to negative gross margin and from 10x P/E to <5x P/E. Even a modest reversion to 35% gross margin and 8x P/E from current levels implies a 60–75% equity drawdown for the memory primaries — without any disorderly cycle. 5) Supply lag is real but not unique. The bullish point about EUV/TSV/hybrid bonding lead times is correct but mis-weighted. The capital cycle history of other capital-intensive industries (oil refining, shipbuilding, semiconductor wafer fab) shows that long lead times increase the eventual amplitude of the down-cycle: capital decisions made at peak are not reversible when conditions soften, leading to capacity overhang. Long lead times delay the down-cycle; they do not abolish it. 6) China is the textbook capital-cycle disruptor. In Chancellor's historical case studies (steel, shipbuilding, solar, panels, batteries), state-backed Chinese entrants repeatedly compressed margins of consolidated Western/Korean/Japanese oligopolies once technology gaps narrowed. The U.S. equipment restrictions on China have created the illusion that this dynamic is paused, but the data shows CXMT doubled DRAM share in 18 months and is targeting domestic HBM3. The structural-change analysis appropriately flagged this; the capital cycle framework would weight it heavier as the single most important multi-year risk. 7) Substitute capital formation is its own supply curve. The capital cycle framework treats financing flows into substitutes as a parallel supply expansion. Cerebras' $5.5B IPO, Marvell's $5B Celestial acquisition, the Sandisk/SK hynix HBF JV, and the CXL ecosystem (ALAB, MRVL, MCHP) are collectively financing "HBM intensity reduction." Even if HBM unit demand is met, the value capture per dollar of AI compute is diluted. Capital is flowing in adjacent to the memory primaries to reduce the share of AI spend that ends up in their P&L. 8) The bull case relies disproportionately on demand visibility. The capital cycle warns against demand-anchored theses. The bull case requires AI capex to continue at current levels or accelerate, hyperscaler ROI economics to remain favorable, sovereign AI to scale, and inference workloads not to migrate to non-HBM architectures. Each of these is plausible; the joint probability that all hold through 2028 is materially lower than the headline narrative suggests. 9) Sell-side estimate trajectory is itself a signal. UBS's PT trajectory ($535 → $1,625, a 3x increase in one revision) is historically associated with peak euphoria. Estimate revisions of this magnitude have a poor forward record. The framework would treat the velocity of estimate revisions as a contra-signal. 10) Where the asymmetry sits. The capital cycle framework reframes the risk/reward calculation. Even if the bull thesis is right and earnings hold through 2028, the upside from current levels is modest (multiple expansion has already happened). If the bull thesis is partially wrong — say, 2028 brings 25% peak-to-trough EPS decline rather than 50% — the equity drawdown is still material because multiples will compress simultaneously. The asymmetry is not favourable at current valuations. Bottom line: The structural change thesis was directionally correct but materially overweighted by the original analysis. The capital cycle framework appropriately reweights toward supply-side caution and treats current peak conditions, peak valuations, peak management confidence, and accelerating capital inflows as a coherent set of late-cycle signals. The memory industry has undergone real and beneficial structural change in shape, but the empirical base rate against the "cycle has been abolished" claim is overwhelming. The economic characteristics of memory businesses have improved but have not been transformed into stable, compounding, low-volatility ones — and the next 18–30 months are statistically more likely to mark the end of this up-cycle than a transition to a new regime.

🇺🇸 A CIA officer got caught with $40 million in gold bars at home. David Rush had a Top-Secret clearance, a senior management position at the CIA, and apparently a very interesting storage situation. Federal agents raided his house last week and walked out with 300 gold bars worth over $40 million, $2 million in cash, and 35 luxury watches, mostly Rolexes. His explanation for the gold? "Work-related expenses." It gets wilder. The man spent nearly 20 years lying about his entire background, fake degrees, a Navy pilot career that never happened, none of it was real. He applied to the CIA three times before finally getting in, adding more fake credentials each time until something stuck. The CIA caught him through an internal investigation and handed it to the FBI. The real question nobody wants to answer is how someone with a completely fabricated resume held Top-Secret clearance for two decades without anyone noticing. Source: NBC NEWS

Elon Musk’s goal for The Boring Company is to solve one of the most miserable daily experiences on Earth: traffic Cities are three-dimensional But transportation is still mostly trapped in a two-dimensional surface network Roads, intersections, bottlenecks, traffic lights, accidents, construction, weather - everything gets stacked on the same flat layer until the entire system chokes The Boring Company’s answer is simple but radical: Go underground Build fast, low-cost tunnel networks under major cities and turn transportation into true 3D infrastructure Right now, the focus is on making tunneling dramatically faster and cheaper with machines like Prufrock, which is designed to mine continuously while installing tunnel liner at the same time But the long-term vision goes much further Local Loop tunnels could move people across cities without surface traffic, while future Hyperloop-style systems could connect entire cities at ultra-high speed Imagine going from Los Angeles to San Francisco, New York to Washington D.C., or Dubai to Abu Dhabi in a fraction of today’s travel time - underground, electric, direct, and protected from surface congestion That is the real mission: Building the missing third dimension of transportation This is how you actually attack soul-destroying traffic at civilization scale