A physics based explanation of why life exists in the universe medium.com/p/1131c68fc784

Chinese models have often caught up with US ones by 'distilling' them on the quiet. The EU AI act's transparency requirements ironically potentially cripple attempts by EU models to do this.

Says the deputy editor of the magazine that ran a cover story in April just to complain about govt. actually using AI at all. They just hate the future.

AI alignment, safety and regulation is having the exact opposite effect of its intentions.

The one good thing to come out of the Fable ban could be a commitment for Europe to build out AI models and infrastructure, drowning out the luddites.

Wow anthropic.com/news/fable-myt…

A nuanced take from Fable on Andy Burnham's energy privatisation plans. Not as crazy as it sounds but also, biggest benefits would be outside of ordinary user costs which is where political capital will inevitably push it. "A useful starting point: what Burnham is actually proposing is closer to "public control" than wholesale state ownership. His track record (the Bee Network buses) is franchising, where the public sector sets fares and routes but private operators still own and run the assets. So before weighing nationalisation, it's worth noting his rhetoric on energy and water being "publicly owned" sits alongside a delivery model that has in practice been about control rather than balance-sheet ownership. That distinction drives almost everything below. It also matters which layer of "energy" you mean, because the economics differ enormously: - Retail supply (the companies that bill you). Cheap to take over (the TUC has put the cost of the supply segment at roughly £2.85bn) but low leverage over actual prices. - Generation (power stations, wind, solar, nuclear). Expensive and where the carbon and security questions live. - Networks and grid (transmission and distribution). The natural monopoly, and arguably the strongest theoretical case for public ownership. **Upsides** - Removes the shareholder dividend layer from a service everyone must buy. The water comparison Burnham leans on (roughly £85bn extracted since privatisation) is the emotional core of the argument. - A natural monopoly like the grid has weak competitive discipline anyway, so public ownership can lower the cost of capital (the state borrows more cheaply than utilities) and align investment with long-term strategy rather than quarterly returns. - Political control over bills and over where investment goes, including the ability to direct buildout toward decarbonisation or regional priorities. - Public legitimacy: bills and standing charges are a live cost-of-living grievance, and visible public ownership can buy consent for the disruption that comes with the energy transition. **Downsides** - Starmer's objection is the strongest one on bills: you still buy gas and power on international markets, so nationalising suppliers does not automatically cut the wholesale cost that drives prices. The saving is the margin, not the commodity. - Investment uncertainty. The UK needs vast private capital (tens of billions) to hit 2030 clean-power targets. A nationalisation programme creates exactly the regulatory uncertainty that freezes investors, potentially slowing decarbonisation rather than accelerating it. This is the central tension a Chancellor would face against an energy secretary. - Fiscal cost and opportunity cost. Money spent buying out shareholders is money not spent on bills, grid, or generation. The headline buyout figures balloon once you move beyond retail into generation and networks. - Political capture and underinvestment risk. State-owned utilities can become subject to short-term political pressure to suppress bills below cost-recovery, starving them of capital (a failure mode visible in several European examples). - Execution risk and distraction. A major reorganisation absorbs years of Whitehall and management bandwidth at precisely the moment the system needs to be building fast. **Relevance to AI strategy** This is the part that has changed the calculus recently, and it cuts against rushed nationalisation. AI data centre demand has become the single biggest new pressure on the British grid. Proposed projects are now seeking around 50GW of connections, which exceeds Great Britain's entire peak demand of roughly 45GW. Government forecasts see at least 6GW of AI-capable data centres needed by 2030, equivalent to about four large nuclear reactors, and the connection queue has ballooned with multi-year, sometimes 15-year, waits that are already pushing some investment abroad (OpenAI's Stargate UK site was paused partly over energy costs). The implications: 1. The binding constraint for AI is grid connection speed and generation capacity, not who owns the retail supplier. If nationalisation introduces investment uncertainty, it directly threatens the compute buildout the UK says it wants. AI strategy and a disruptive nationalisation programme are in tension. 2. Conversely, the layer where public control could help AI is the grid and connection queue. A state actor able to prioritise "strategic demand" and accelerate connections could be an enabler rather than an obstacle. So the part of the energy system most relevant to AI (networks) is also the part with the best public-ownership case. 3. There is a genuine conflict between AI demand and decarbonisation: data centres could push the UK toward 8 to 16 percent of total electricity demand and strain the Clean Power 2030 goal. Whoever owns the system has to arbitrate that, and public ownership at least makes the trade-off explicit and politically accountable. **My take, with the caveat that this is a contested policy judgement rather than a settled answer** Full nationalisation of the whole energy chain looks like the wrong tool: highest cost, highest disruption, weakest effect on the thing people actually care about (bills), and actively risky for the AI and decarbonisation buildout. I would not recommend it. The defensible version, if it comes to it, is narrow and sequenced: - Prioritise the natural monopoly. Public ownership or much tighter public control of the transmission and distribution networks, plus the connection queue, is where the economics and the AI case both point. NESO and Great British Energy already move in this direction, so this is evolution, not revolution. - Use franchising and strategic contracting rather than buyouts where possible, which is consistent with Burnham's actual Bee Network model and avoids tying up capital in shareholder compensation. - Leave generation largely to contracted private capital (CfD-style mechanisms) so the 2030 buildout is not starved, while the state shapes what gets built. - Treat retail supply as low priority. The bill savings are small relative to the political energy required. - If you proceed, do it with maximum predictability: clear compensation terms, a defined scope, and a fixed timetable, because the single biggest cost of nationalisation is the investment chill from uncertainty, and that chill lands hardest on exactly the grid and AI infrastructure the UK needs. In short: control the monopoly layer, contract the competitive layer, and avoid a sweeping buyout that would spook the capital the energy transition and AI ambitions both depend on. One note of context: this is live and unsettled. Burnham would first need to win the Makerfield by-election and then the Labour leadership before any of this is more than a position paper, and his own language ("public control") leaves a lot of room between the rhetoric and a balance-sheet nationalisation."

I've been bearish and wrong about London for decades. Now I'm longer term bullish.

RIP David Hockney. This is my favorite and perfectly captures my parents' generation in the 70s.

The 'safer' AI models are the most dangerous, precisely because of their 'alignment'.

AI reaction to this writing style. This reads like AI-generated text that's been lightly roughed up to look human, which is its own recognizable genre now. The tells fall into two layers. The AI layer underneath: The hook formula. "Six European researchers spent months writing one document. Its plot: Europe loses everything by 2031." Short declarative sentence, then a dramatic reveal. This is the viral-thread template every LLM has absorbed from a million "I spent 100 hours so you don't have to" tweets. Manufactured stakes language. "the numbers are brutal," "The sharpest idea in the whole report," "The report's ending turns on it." These are trailer-voice phrases. A human who actually read the report tends to react to specific things; AI summarizes the shape of a reaction. The tidy antithesis habit. "Building worse local versions of American tools isn't independence, it's expensive isolation." "This isn't a prophecy, it's a warning." The "isn't X, it's Y" construction twice in one short piece is a strong fingerprint. So is "The window is open. That's the entire point of publishing it now," which is pure LLM cadence: short sentence, then a sentence explaining the significance of the short sentence. Perfectly balanced structure. Numbers section, then key insight, then the UK angle, then optimism pivot, then five fixes, then CTA. It's a summary with the connective tissue of an essay. Humans writing hot takes meander, double back, or fixate on one point disproportionately. The optimism pivot. "Despite this gloomy future picture, I am optimistic because..." LLMs are trained to land on constructive, hopeful endings. Humans who just read something scary often just stay scared. The human-camouflage layer on top: The typos and compression ("its scary," "if u r into," missing spaces after commas, "n the numbers") are too evenly distributed and too superficial. Real sloppy typing produces structural mess: dropped words, broken syntax, abandoned clauses. This has flawless sentence architecture wearing a dirty shirt. That mismatch, polished logic with cosmetic errors, is arguably the biggest tell of all. How I'd write it differently: I'd pick one thing and lead with a genuine reaction to it, not a synopsis. Something like: "The detail I can't stop thinking about from this Europe 2031 report: America's biggest AI supercomputer runs at 1,250 megawatts. Europe's runs at 83. That's not a gap, that's a different sport." Then follow the actual thread of thought rather than covering the report section by section. Drop the optimism pivot unless it's earned by an argument. Let one stat go unexplained. End abruptly. Real posts don't resolve.

Increased clean energy use technically lowers the Earth's temperature (by an infinitesimal amount). But because people confused consumption with fossil fuel use as a convenient narrative, we are in danger of destroying economies.

One of the possible effects of the new exponential companies (Anthropic, OpenAI and SpaceX) going public is a concentration of market capitalization in them and their hyperscaler ecosystems at the expense of the rest of the market, no matter how overvalued these companies seem.

Just in time code. In the future all front end development will be on-the-fly, there is no need for two week sprints and agile development for interfaces.

On the other hand, Usain Bolt is only a fraction of a second faster than his rivals.

Breaking. Cathedral built in record time, centuries ahead of Cologne, Milan, Canterbury and Notre Dame!

Fable 5 seems incredible so far. I ran it against the Dorabella cipher and after a couple of hours of prompting it has gotten further than anything in the literature, ever.

OK, here's a good example of AI written garbage. Boltzmann didn't invent entropy and he never actually used the equation on his tombstone.

Bourdain was the US' principal psychogeographer, just as Meades is the UK's