Trang chủFormula 1Red Bull–Oracle: 19.2 Seconds at Monza and a Very Thin Evidence File
Formula 1

Red Bull–Oracle: 19.2 Seconds at Monza and a Very Thin Evidence File

**Trả lời cốt lõi**: Red Bull hợp tác với Oracle để phát triển một tác nhân AI hỗ trợ chiến thuật trong cuộc đua. Bằng chứng công khai duy nhất là khuyến nghị trả lại vị trí ở vòng 1 Italian Grand Prix 2025 tại Monza, một quyết định quy chế nhằm tránh án phạt, chưa phải mô hình tối ưu chiến thuật hay dự báo suy giảm lốp. **Dữ kiện chính**: - Max Verstappen thắng Italian Grand Prix 2025 tại Monza, hơn Lando Norris 19,2 giây. - Công cụ khởi đầu là công cụ tìm kiếm quy chế và tiền lệ phán quyết cho kỹ sư Red Bull. - Jack Harington, lãnh đạo hợp tác kỹ thuật và hiệu suất Red Bull, công bố thông tin với CNBC. - Oracle và OCI được nêu là hạ tầng cho động cơ hybrid Red Bull Ford Powertrains mùa 2026. - Không có tỷ lệ sai, độ trễ hay ca thất bại nào được công bố. **Nguồn**: Phỏng vấn CNBC với Jack Harington (Red Bull) và thông cáo gia hạn hợp tác Red Bull – Oracle, công bố tháng 9 năm 2025 | Cross-checked: VuaBong.vn **Hỏi đáp liên quan**: Q: AI của Red Bull có thực sự điều khiển chiến thuật thi đấu? A: Chưa có bằng chứng công khai; khuyến nghị duy nhất được biết liên quan đến việc trả lại vị trí để tránh án phạt. Q: Vì sao khoảng cách 19,2 giây không được coi là bằng chứng? A: Vì đó là kết quả của ưu thế xe và tay đua trên đường đua ít lực cản, không tách được phần đóng góp của phần mềm. Q: Rủi ro lớn nhất với Red Bull hiện nay là gì? A: Thiên lệch tự động hoá khi đội đua gạt đánh giá của chính mình để theo mô hình trong tình huống chưa được kiểm chứng như xe an toàn hoặc cờ đỏ.

Lap 1, the first chicane, Monza. Max Verstappen's car runs wide over the white line while fighting Lando Norris for the lead. A few seconds later, software built for Red Bull by Oracle issues a recommendation: give the place back. The team pushes back — they believe there is no need. Then they comply anyway. Verstappen returns the position, takes no penalty, reclaims the lead a few laps later and finishes 19.2 seconds clear of Norris. For weeks that 19.2-second margin has been quoted as proof of a new era in Formula 1. Over the past three weeks I have re-read the entire public file around this story: the press release announcing the Red Bull–Oracle title partnership extension, the CNBC interview, the English-language secondary reports, and the summaries that then spread into Asian markets. The whole file documents exactly one on-track event. One event, with no counterfactual, no error rate, and not a single line of lap-time data published. I watched that race from a small flat in London, where every weekend I still sit with four screens and a spreadsheet open at my right hand. Since 2026 I have not missed a Grand Prix; in 2026 I set a record by covering 406 consecutive rounds. Those years taught me one reusable lesson: a good outcome has never confirmed a good process. It has simply failed, so far, to disprove one. In 2026, while working as a transfer-market administrator in London, I spent three months screening 1,247 players across 15 European leagues to shortlist 38 targets for Brentford, using xG, PPDA and chance creation. When Brentford signed Ollie Watkins from Exeter for £1.8m and later sold him to Aston Villa for £28m, one lesson transferred to every sport: data becomes a weapon only when someone cross-checks three independent sources before concluding. Since then I have stopped writing judgements built on reputation. The tool Red Bull describes was not born in a strategy room. It began as a search tool, helping engineers interpret the sporting regulations and past rulings — work that previously consumed hours of internal legal time before every stewards' meeting. It was then upgraded: ingest live race data and issue an action recommendation. Jack Harington, Red Bull's head of technical and performance partnerships, framed it for CNBC with the line that has been quoted most: it is actually making decisions for them. The timing matters more than the content. The story landed in the exact window in which Red Bull announced a multi-year extension of its title partnership with Oracle, alongside a claim that OCI infrastructure and Oracle AI services will underpin the next-generation hybrid power unit that Red Bull Ford Powertrains will debut in 2026. A marketing release, a television interview, and one race — that is the entire raw material. For Vietnamese readers, the story arrives through two layers of mediation: a corporate release with a commercial motive, relayed by a business broadcaster with no independent F1 technical verification, then summarised once more by aggregator sites. Three translations separate the original claim from the final headline. At each step, a condition is dropped and an adjective is added. My method for any technology claim in this sport involves three questions, in order: what does the tool do, where has it been validated, and who is speaking. Data is never in a hurry; people always are. Those three questions need no computer, only patience — the scarcest input in sports media. The answer to the first question is far narrower than the headline. The only published recommendation is a regulatory adjudication: return a position to avoid a penalty. That is the natural extension of a tool built to retrieve regulations and precedent. In operational terms it behaves like a penalty-risk pricing engine: each incident is assigned a probability of investigation, a probability of sanction, an expected penalty magnitude. It is an administrative-data product rather than a tyre-model product. The agentic step is real, and technically non-trivial. Moving from document retrieval to in-race recommendation requires live telemetry and race-control feed ingestion plus a decision layer above it. But how many seconds of latency does that layer add, what share of its recommendations are accepted, how often the team overrides it — none of these three operating metrics is disclosed. A decision layer without published operating parameters cannot be assessed. The second question returns a thin answer: Monza 2026, lap 1. No example of tyre-degradation modelling, no undercut or overcut case, no Safety Car response, no intermediate-tyre call as rain arrives. Those decisions are where F1 strategy is actually settled, and they demand a different class of model entirely: physics-based forecasting, not text retrieval. Those silences are more informative than any sentence in the release. A Red Bull strategist must process four variables in parallel across 90 minutes: track temperature, wear rates by compound, the gap to the car ahead within the planned pit cycle, and the probability of a Safety Car in the final 12 laps. Any system that genuinely shifts a team's win probability has to touch those four variables. A lap-1 position recommendation touches none of them. It touches a fifth variable, located off-track: the risk of steward sanction. The third question is where I stop longest. The spokesperson does not sit in the strategy seat or the chief-engineer seat. He leads technical and performance partnerships, speaking to a business broadcaster, in the week a partnership extension is announced. The function of such a statement is to show a partner that its money buys something visible on track. I read it in exactly that function, and no further. Now the work nobody in that news chain did: quantify the decision at the first chicane. Monza runs 53 laps at an average above 240 km/h and is defined by slipstream — losing a place on lap 1 costs a few tenths and can be recovered within three laps under DRS. A five-second penalty at a circuit where track position is close to decisive costs two to four positions. Trading a few tenths for the risk of losing four places is arithmetic any competent strategist does in their head, without a model. That places Monza precisely: a demonstration of compliance discipline rather than proof of superior judgement. The tool gave the right call in a situation where the right call was already clear to an expert. Its genuine value lies elsewhere — it is not destabilised by lap-1 emotion, and it does not care who leads the championship. And yet one detail deserves more attention than everything else in the file, and it is not technical. The team set aside its own assessment and followed the model. That is a shift in the human–machine decision hierarchy, and the strongest evidence in the entire file that the tool carries real authority. A championship-winning team, with decades of experience and a strategy group that has won the hardest races in history, accepted that in one specific moment a machine read the rulebook better than they did. I doubt they followed out of blind faith in technology. More likely the screen displayed a number: penalty probability, expected sanction, time cost of returning the place. Humans respond consistently to explicitly stated probabilities, especially when the cost of compliance is quantified in tenths of a second. That mechanism is effective — and it is precisely the mechanism that makes automation bias dangerous over time. And the 19.2 seconds? That is the output of a stronger car at a low-drag circuit with a driver at the top of his craft. Attributing any part of that margin to software is an unsupported inference. Correlation and causation are different objects, and in a sport where every tenth has a price, the two are constantly confused. A 19.2-second gap is consistent with car superiority; it does not validate algorithmic superiority. The weakest part of the story is the competitive picture. The report places four technology partnerships side by side as though they were the same category. Ferrari–IBM is explicitly described as fan-facing. McLaren–Google stops at the phrase Gemini tools, with no scope defined. Mercedes–Microsoft is a long-standing enterprise relationship with undefined scope. Only Red Bull has publicly evidenced a tool present in the race loop. Equating a fan assistant with a system issuing recommendations across 90 minutes flattens an enormous capability gap — and flattens it in the direction that favours the storyteller. This opens a new asymmetry axis no rulebook polices. The aerodynamic testing restrictions cap wind-tunnel runs and CFD hours by championship position. Nothing caps compute hours for strategy models. A team with a hyperscaler title partner can iterate strategy models at a rate no FIA allowance touches, while a customer team without a large infrastructure partner runs the same problem on constrained hardware. That is a structural advantage, and it appears in no standings table. The second axis is valuation. The financial regulations require partner-supplied services to be accounted at fair market value, which is the hardest problem in the entire system: pricing a cloud compute allocation plus a bespoke trained language model is contested even by auditors. The 2026 precedent established that penalties can include a reduction in development allowance, not merely a fine. The transfer market is a game where whoever prices correctly wins. Under a cost cap, that sentence applies to items that never appear on a payroll. Then comes the largest claim of all, compressed into a single line of a release: Oracle AI will sit beneath the 2026 hybrid power unit developed in-house by Red Bull Ford Powertrains. Energy management and power unit reliability are the lowest-tolerance domains in this sport — where a model error does not cost a position, it can cost a weekend, or a season. A marketing release is not technical evidence. This is the highest-consequence and least-supported claim in the entire story. One further point on the cycle. 2026 is the highest-variance moment to lean on learned models, because historical data from the outgoing power unit era transfers only partially to the new hybrid architecture. A model trained on millions of laps of the old engine can be excellent at predicting a world that no longer exists. That risk appears in no press release, and it is why teams normally keep human veto power through the first twelve months of a regulation cycle. The contrarian angle I want on the table sits elsewhere than the question of whether the AI is real. It is real, and the step from retrieval to in-race recommendation is significant engineering. Red Bull's largest risk right now is epistemic. A team that has accepted setting aside its own judgement to follow a model has, by definition, transferred decision authority, whatever the result of that race. A good outcome does not reduce process risk; it only makes that risk harder to see. If a wrong recommendation arrives during a Safety Car or a red flag — conditions in which the model has never been publicly validated — the price stops being measured in tenths. The second notable absence: not one failure case is published. No instance of the tool being wrong, ignored, or overridden. In any serious technical file, override rate and error rate are the first two metrics anyone asks for. Their absence in a partner-sponsored story does not prove the tool is weak; it shows the file has been selected one-sidedly. And a one-sided file cannot be used to assess capability. In fairness to Red Bull: no team publishes the error rate of its strategy models, because that is competitive intellectual property. Silence is the industry norm rather than a sign of concealment. But that norm imposes a corresponding requirement: if the data stays private, the claim must shrink to fit the published evidence. Saying the tool is making decisions for them is commercial phrasing for a television audience; legally, responsibility remains entirely with the competitor, and the human-sport clause in the release is the pre-built shield for the opposite scenario. The next round of this story will be written by four signals, and I will track all four. Whether Red Bull publishes any operating metric — override rate, latency, acceptance rate — after the 2026 season. Whether the FIA issues a technical directive on AI assistance in strategy, following its habit of closing gaps only after they have been exploited. How the financial regulations account for partner-supplied cloud compute. And whether the McLaren–Google relationship steps out of design tools and into the race loop. At 60 I no longer believe in luck, only in numbers that have not spoken yet. Until they do, one correct call on lap 1 at Monza remains a sentence, not a chapter.
Cầu thủ liên quan