DeepSeek 再被懷疑用 Google Gemini 訓練新版 R1 模型
其中一個支持這論點的是澳洲開發者 Sam Paech，他在 X 上發文指出R1-0528 模型的語言風格與 Google Gemini 2.5 Pro 極為相似。他認為 DeepSeek 已經從以往基於 OpenAI 的數據切換至 Gemini 的合成數據。另一位開發者 SpeechMap 則發現，R1 模型生成的'推理痕跡'（AI 在得出結論時的思維過程）也與 Gemini 模型極為相似。
If you're wondering why new deepseek r1 sounds a bit different, I think they probably switched from training on synthetic openai to synthetic gemini outputs. pic.twitter.com/Oex9roapNv
— Sam Paech (@sam_paech) May 29, 2025
另一邊廂非牟利 AI 研究機構 AI2 的 AI 專家 Nathan Lambert 更發文指 DeepSeek 在缺乏 GPU 和鉅額資金的支持下，也一定會透過市場最佳的模型 API 來蒸餾數據，這次就是 Gemini。
2024 年時，OpenAI 透過金融時報發聲，指他們獲得證據指 DeepSeek V3 是透過蒸餾 ChatGPT 的數據來訓練而成，後來 Bloomberg 也報道指主要金主 Microsoft 偵測到在 2024 年年底，有大量資料經過 OpenAI 開發者帳戶外洩，他們相信是與 DeepSeek 有關。
為防止競爭對手利用其模型數據，AI 公司正加強安全措施。例如，OpenAI 現在要求用戶完成身份驗證才能訪問高級模型，而 Google 則開始對 Gemini 模型生成的'推理痕跡'進行摘要處理，讓競爭對手更難以利用其數據。
更多內容：
DeepSeek may have used Google's Gemini to train its latest model
DeepSeek 懶人包｜中國AI新創如何影響美國AI巨企？一文整理歷史、最新影響及未來
中國 DeepSeek AI 模型自稱 GPT-4，「AI 天材」是抄襲還是幻想？
DeepSeek 反客為主！連百度搜尋都已確定引入
緊貼最新科技資訊、網購優惠，追隨 Yahoo Tech 各大社交平台！
🎉📱 Tech Facebook：https://www.facebook.com/yahootechhk
🎉📱 Tech Instagram：https://www.instagram.com/yahootechhk/
🎉📱 Tech WhatsApp 社群：https://chat.whatsapp.com/Dg3fiiyYf3yG2mgts4Mii8
🎉📱 Tech WhatsApp 頻道：https://whatsapp.com/channel/0029Va91dmR545urVCpQwq2D
🎉📱 Tech Telegram 頻道：https://t.me/yahootechhk
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Android Authority
28 minutes ago
- Android Authority
Worried about the Pixel 10 Pro XL benchmark controversy? Here's why you shouldn't be
Well, friends, the day has finally come: Google Pixel day. Today is the day that Google will officially reveal the Pixel 10 series after months of leaks, and it's a day we should all be excited about. Instead, a lot of Pixel fans are upset and complaining. Those complaints stem from a last-minute Pixel 10 Pro XL benchmark leak, supposedly showing just how fast Google's new Tensor G5 chip is. Assuming the benchmarks are legit, the good news is that the Tensor G5 is faster than the Tensor G4. However, the G5 still trails behind Qualcomm's Snapdragon 8 Elite — and that's what has some people furious. Redditors have called this 'pathetic,' 'sad,' 'a f***ing joke,' and so on. But I've got to be honest — I couldn't care less about any of this. And for most of you reading this, you shouldn't care either. What do you think of these leaked scores if true? 1395 votes I'd be happy with these scores 7 % It's okay but could be better 17 % I'd be disappointed 44 % I don't care about benchmarks 32 % Why you shouldn't read too closely into these benchmarks As we head into today's Made by Google event, the Tensor G5 chip is almost a bigger deal than the Pixel 10 phones themselves. We expect the G5 to be Google's first 3nm chip and the first Tensor chip manufactured by TSMC rather than Samsung Foundry. Those two 'firsts' should translate to numerous upgrades over the Tensor G4, performance being one of them. And as we see from these benchmarks, it looks like that's what we're getting. If we take these leaked benchmarks at face value, the Pixel 10 Pro XL has almost a 22% increase in single-core CPU performance compared to the Pixel 9 Pro XL, along with a staggering 46% increase in multi-core performance. That's an impressive year-over-year upgrade, and it's one I just can't be upset about. But that's clearly not how everyone feels. Whether it's the fact that these GeekBench scores are still lower than Snapdragon 8 Elite phones or that the Pixel 10 Pro XL's GPU scored lower than the Pixel 9 Pro XL, a lot of folks are already ready to write the Tensor G5 off even before they've used the chip themselves. Joe Maring / Android Authority I think there are two ways to view this, both of which make this complaining look a little ridiculous. On the one hand, let's say these benchmarks are 100% accurate. If that's the case, I struggle to see what the big deal is. I spent a lot of time with the Pixel 9 Pro last year (and more recently the Pixel 9a), and performance with the Tensor G4 chip on both phones is great. The UI is as smooth as butter, apps open and run quickly, and the few games I play run without a hitch. If I'm solely looking at benchmark numbers or closely analyzing frames per second, I'm sure I'd be a lot more disappointed with the G4. But for day-to-day performance with the way I use my phone, not once have I felt that the Tensor G4 is underpowered. Seeing that the Tensor G5 could be almost 50% faster for certain tasks is pretty incredible — regardless of how it compares to Qualcomm's chip. Benchmarks — especially for unreleased phones — never tell the whole story. The other way to look at this benchmark leak, which is perhaps the healthiest way, is that benchmarks — especially for unreleased phones — never tell the whole story. Benchmark results are one of the easiest things to fake, so wholeheartedly trusting leaked results like this is never a good idea. And even if the benchmarks are real, there's a very strong possibility that old drivers and/or unoptimized software are leading to these 'disappointing' results (which could very well explain the lower-than-expected GPU Antutu scores). Add all of this together, and it's hard to see this outrage as anything but complaining just for the sake of it. Don't want to miss the best from Android Authority? Set us as a preferred source in Google Search to support us and make sure you never miss our latest exclusive reports, expert analysis, and much more. Find out more here. There are other reasons to be excited about the Tensor G5 Robert Triggs / Android Authority Looking beyond the benchmark debacle, it's also worth mentioning that performance is just one of the reasons to look forward to the Tensor G5. And, if you ask me, it's the least important. While the Tensor G5 should be a faster chip than the G4, it should also be more efficient — and that's what could matter the most. Although I've never had any serious performance issues with Tensor chips, I have had my fair share of complaints about battery life and thermal management, both of which could be resolved with more efficient silicon. We live in a world where flagship Qualcomm chips (when paired with a large enough battery) can easily last a day and a half or two days per charge — something I've never experienced with a Pixel. We also expect competing flagship chipsets to handle basic daily tasks without overheating, something Google's Tensor chips have always struggled with. It's these potential upgrades, rather than faster performance, that have me the most excited for the Tensor G5. I certainly won't say no to a faster phone, but if the G5 enables the Pixel 10 series to have genuinely great battery life and no thermal issues, that's what will make the most significant difference — and that's not something you'll see in any CPU or GPU benchmark. Of course, whether the G5 delivers on this potential remains to be seen until we get our hands on the Pixel 10 lineup, but that again reinforces my point here. One benchmark leak can only tell us so much about a phone. It doesn't reveal anything about real-world performance, battery life, thermals, etc. I can understand being disappointed about 'bad' benchmark results, but to be enraged about a leak like this just doesn't make sense. The good news is that the Pixel 10 is being announced today, so it won't be long now until we can do that testing ourselves to see how the Tensor G5 performs in the real world — and we can (hopefully) put this silly benchmark complaining behind us. Follow
Business Insider
29 minutes ago
- Business Insider
OpenAI engineer's advice to high school students? 'Absolutely learn to code'
No, AI doesn't mean you don't have to learn to code, at least according to one OpenAI researcher. Szymon Sidor recently sounded off on why, even in a world with ChatGPT, Cursor, and other AI tools, it's still good for current high school students to learn basic coding. "You should absolutely learn to code," Sidor said during episode of the company's podcast from August. "One skill that is at a premium and will continue to be at a premium is having really structured intellect that can break down complicated problems into pieces." Sidor, who has worked at OpenAI for over eight years, said programming is a great way to learn such a skill. Andrew Mayne, who hosts the podcast and previously served as OpenAI's chief science communicator, questioned the notion that people would downplay learning to code simply because AI tools can do it for you. "Whenever I hear people say, 'Don't learn to code,' it's like, do I want an airplane pilot who doesn't understand aerodynamics? This doesn't make much sense to me," Mayne said. Their views echo some other top voices in the industry, including GitHub CEO Thomas Dohmke. Dohmke said in April that coding should be taught as a core subject. AI's coding abilities continue to advance. Meta CEO Mark Zuckerberg told Joe Rogan earlier this year that the company's AI would soon be able to write code that matches the output of a mid-level engineer. For his part, OpenAI CEO Sam Altman previously said his biggest piece of advice for high schoolers was to master using AI. "The obvious tactical thing is just get really good at using AI tools. Like when I was graduating as a senior from high school, the obvious tactical thing was get really good at coding," Altman told Stratechery in March. "And this is the new version of that."
Yahoo
an hour ago
- Yahoo
Trending tickers: Nvidia, Palantir, Oracle, Strategy and Convatec
Nvidia (NVDA) Tech stocks sold off on Tuesday, with chipmaker Nvidia (NVDA) closing the session 3.5% in the red, and hovered just below the flatline in pre-market trading on Wednesday. The US tech-focused Nasdaq (^IXIC) ended Tuesday's session down around 1.5%, and futures linked to the index (NQ=F) were down 0.3% at the time of writing. Read more: Stocks slip as UK inflation jumps to highest level since January last year The fall in tech stocks came after a report by the Massachusetts Institute of Technology's Nanda initiative said that "95% of organisations are getting zero return" from their investments in generative artificial intelligence (AI). The sell-off also followed a report by The Verge on Friday which said OpenAI CEO Sam Altman had pointed to an AI bubble. Palantir (PLTR) Data software platform provider Palantir (PLTR) slid more than 9% on Tuesday, and was down a further 1.4% in pre-market trading on Wednesday. The fall on Tuesday marked a fifth straight losing session for Palantir shares. However, the stock is still up nearly 109% year-to-date, having risen more than 150% from its April low thanks to its second quarter earnings report, which saw the company's revenue top $1bn in a single quarter for the first time. Stocks: Create your watchlist and portfolio Steve Clayton, head of equity funds at Hargreaves Lansdown, said the fall in Palantir shares came as "investors questioned whether a valuation of over $400bn could be justified by a business which last year generated less than $3bn of revenues". The stock also came under pressure after a bearish report by short seller Citron Research published on Monday predicted a price target of $40. The firm's founder, Andrew Left, called the target "generous." Oracle (ORCL) Cloud technology company Oracle (ORCL) was another company in the sector that fell on Tuesday, sliding 5.8% and lingering just below the flatline in pre-market trading on Wednesday morning. It followed a Bloomberg report on Monday which said Oracle's longtime chief security officer Mary Ann Davidson was leaving the company as part of a recent reorganisation. A spokesperson for Oracle had not responded to Yahoo Finance UK's request for comment at the time of writing. Strategy (MSTR) Cryptocurrency-related stocks fell on Tuesday, as the price of bitcoin (BTC-USD) declined after hitting a fresh high last week. Bitcoin hit a record $124,000 last week, but has since fallen back, dipping 3% over the past five days. Read more: UK inflation rises to 18-month high in July, driven by higher air fares This has weighed on crypto-related stocks, including software company Strategy (MSTR), which is one of the largest corporate holders of bitcoin. Strategy shares tumbled more than 7% on Tuesday, though the stock was trading 1.3% in the green in pre-market trading on Wednesday. Convatec (CTEC.L) On the London market, Convatec (CTEC.L) was the biggest riser on the FTSE 100 (^FTSE) on Wednesday morning, surging more than 7%. The rise in shares came after the medical products and technology company announced that it was commencing a $300m share buyback programme. Convatec said that the programme would start immediately and would run until 31 December, adding that it could be extended. Read more: How rising inflation will affect your state pension Should CEO pay be capped? Have your say How to find the weak link in your financesError in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
