The chart and the scaling-wall debate
->Diagrammet og skalavæg-debatten
In November 2024, media reported that OpenAI's next model, code-named Orion, showed only modest gains over GPT-4. That revived fears that the long-rumored AI scaling wall — the idea that throwing more data and compute at models yields diminishing returns — had finally arrived. OpenAI CEO Sam Altman pushed back, posting that there was no wall. Musk's chart now suggests the opposite: that after a brief lull, progress has resumed at a pace that dwarfs anything before.
Danish:I november 2024 rapporterede medierne, at OpenAI's næste model, med kodenavnet Orion, kun viste beskedne forbedringer i forhold til GPT-4. Det genoplivede frygten for, at den længe omtalte AI-skalavæg — ideen om, at at kaste mere data og beregning efter modeller giver faldende afkast — endelig var nået. OpenAI's CEO Sam Altman afviste det og skrev, at der ikke var nogen væg. Musks diagram antyder nu det modsatte: at efter en kort pause er fremskridtene genoptaget i et tempo, der overgår alt tidligere.
Next:The chart itself is unlabeled beyond the November 2024 marker and the spring 2026 breakout. It does not specify which metric — benchmark scores, parameter counts, or something else — is being plotted. Musk offered no source for the data, and the accuracy of the trajectory through the rest of 2026 remains unverified.
Danish:Diagrammet selv er umærket ud over november 2024-markøren og foråret 2026-gennembruddet. Det specificerer ikke, hvilken metrik — benchmark-score, parametertal eller noget andet — der afbildes. Musk gav ingen kilde til dataene, og nøjagtigheden af kurven gennem resten af 2026 er fortsat ubekræftet.
Next:Musk's own AI bets
->Musks egne AI-satsninger
Musk's xAI has been making moves of its own. Grok 4.5 recently topped an independent agent benchmark from Artificial Analysis, beating rivals on both cost and speed. Musk also revealed that SpaceX engineering data — excluding material restricted under U.S. arms-export rules — will feed Grok's next training run. The idea is to improve the model's reasoning on physical problems, a domain where current AI often stumbles.
Danish:Musks xAI har også gjort sine egne tiltag. Grok 4.5 toppede for nylig en uafhængig agent-benchmark fra Artificial Analysis og slog konkurrenterne på både pris og hastighed. Musk afslørede også, at SpaceX-tekniske data — eksklusive materiale, der er begrænset under amerikanske våbeneksportregler — vil indgå i Groks næste træningsrunde. Ideen er at forbedre modellens ræsonnement om fysiske problemer, et område, hvor nuværende AI ofte fejler.
Next:In a notable reversal, Musk called Anthropic 'the current industry leader' after months of public criticism. The comment suggests a shifting landscape where even Musk acknowledges competitors even as he pushes his own models forward.
Danish:I en bemærkelsesværdig vending kaldte Musk Anthropic for 'den nuværende industrileder' efter måneders offentlig kritik. Kommentaren antyder et skiftende landskab, hvor selv Musk anerkender konkurrenter, selvom han skubber sine egne modeller fremad.
Next:Timelines and disagreements
->Tidslinjer og uenigheder
At Davos in January 2026, Musk predicted that artificial general intelligence — AGI — could surpass one person's intelligence before the end of 2026. He separately projected that AI will exceed all of humanity's combined intelligence within five years. Musk frames AI as both a historic economic opportunity and an existential risk, warning it could slip beyond human control.
Danish:Ved Davos i januar 2026 forudsagde Musk, at kunstig generel intelligens — AGI — kunne overgå en persons intelligens inden udgangen af 2026. Han forudsagde separat, at AI vil overgå hele menneskehedens samlede intelligens inden for fem år. Musk rammer AI ind som både en historisk økonomisk mulighed og en eksistentiel risiko og advarer om, at den kan glide ud over menneskelig kontrol.
Next:Not everyone buys the accelerated timeline. Computer scientist Geoffrey Hinton, often called the godfather of deep learning, disagrees. Hinton stated that broad AGI could still be up to two decades away. Ben Goertzel, CEO of the Artificial Superintelligence Alliance, offered a middle view: advanced AI systems already rival state-level capabilities, lending some support to Musk's sense of urgency even if the exact dates are off.
Danish:Ikke alle køber den accelererede tidslinje. Computervidenskabsmanden Geoffrey Hinton, ofte kaldt deep learning's gudfar, er uenig. Hinton har udtalt, at bred AGI stadig kan være op til to årtier væk. Ben Goertzel, CEO for Artificial Superintelligence Alliance, gav et mellemliggende syn: avancerede AI-systemer matcher allerede statslige kapaciteter, hvilket giver en vis støtte til Musks følelse af hastende karakter, selvom de nøjagtige datoer er forkerte.
Next:The debate is not academic. If Musk's chart is right, the next few years will see capabilities that make today's models look primitive. If Hinton is right, the industry has years of steady improvement ahead. Either way, the conversation has moved from 'if' to 'when.'
Danish:Debatten er ikke akademisk. Hvis Musks diagram er korrekt, vil de næste par år
