Latest news with #DeepSeek
Mint
44 minutes ago
- Business
- Mint
‘DeepSeek moment' in AI vs humans: Artificial intelligence influencers outperform human rivals in livestream sales
Artificial intelligence (AI) avatars are now selling more than real people. A live stream hosted by Chinese tech firm Baidu and popular live streamer Luo Yonghao confirms this. On Baidu's platform, Youxuan, Luo and his co-host Xiao Mu used digital versions of themselves to livestream for over six hours. Their AI avatars helped them earn 55 million yuan (about $7.65 million), CNBC reported. This was much more than Luo's earlier livestream using his real self. The earlier stream lasted four hours and made fewer sales. Luo admitted it was his first time using virtual human technology. 'The digital human effect has scared me ... I'm a bit dazed,' CNBC quoted him as saying. Luo became famous through livestreaming on Douyin (China's TikTok). He has over 24 million followers. He started selling online in 2020 to pay off debts from his failed phone company. The avatars were built using Baidu's AI model, which was trained on five years of videos and copied Luo's humour and style. China's livestreaming and digital avatar industry is growing fast. AI company DeepSeek is gaining attention for creating technology like ChatGPT, but at a lower cost. DeepSeek uses open-source tools and is backed by Chinese tech giant Baidu. AI-powered digital avatars are helping companies save money. They can livestream nonstop, without needing breaks or big production teams. Earlier, companies like Baidu were unsure about using digital humans. But, today, their technology has improved greatly. Since the pandemic, livestream shopping has become popular in China as people looked for new ways to earn money. Livestreamers now earn through commissions and digital gifts. The trend is so strong that Douyin became the second-biggest online shopping platform in China. It has overtaken and competing with market leader Alibaba. 'This is a DeepSeek moment for China's entire livestreaming and digital human industry,' Wu Jialu told CNBC. Wu is the head of research at Be Friends Holding, another company owned by Luo. Neuro-sama, an AI streamer, is gaining popularity on Amazon's platform Twitch. She appears as an anime girl who chats, sings and plays games like Minecraft. Her replies come from a chatbot similar to ChatGPT. Her creator, Vedal, built her using the Unity game engine and works on this project full-time, according to Bloomberg. Despite being an AI avatar, Neuro-sama has become a hit. Around 5,700 viewers watch her streams regularly, making her one of Twitch's top streamers.
Yahoo
12 hours ago
- Business
- Yahoo
19 People Becoming Billionaires In The AI Boom
Artificial intelligence (AI) is perhaps the most significant technological step forward since the introduction and proliferation of social media networks and ecommerce platforms, both in terms of its impact on the business world as well as society writ large. Read More: Find Out: It's no secret that the rapid rise of AI models — and the companies which either build, provide hardware to power, or simply leverage these tools in the marketplace — has led to a massive creation of capital. Who, exactly, are the most prominent new billionaires to profit from the AI boom? Forbes provided a few details outlining these personalities. Perhaps the most prominent of these newly minted AI billionaires, Alexandr Wang is the founder and CEO of Scale AI, a company which engaged in data labeling for other tech juggernauts such as OpenAI, Google and Meta. His current net worth was pegged at approximately $2 billion as of April, per Forbes. After dropping out of MIT at the age of 19, Wang is now poised to make another controversial move. According to TechCrunch, Wang received a very enticing offer to join Meta's AI team as the social media company redoubles its efforts to build a 'superintelligence' capable of dethroning competitors in the AI arms race. Founder and CEO of Chinese AI company DeepSeek, Wenfeng's firm rose to fame as a disruptor in the AI space, making headlines for months earlier this year due to the release of its free artificial intelligence tools which were said to rival paid options coming from OpenAI, among others. Wenfeng's estimated net worth as of April was about $1 million. Perhaps best known as CEO of one of the world's most famous tech companies, Google, since 2019, Pichai was in the driver's seat for Google's entry into the AI revolution via the release of its Gemini model. With Google shares having enjoyed a rapid increase in valuation — and despite the fact that Pichai owned just 0.02% of the company's total shares in April — at that time his net worth was pegged at around $1.1 billion. Splitting from OpenAI to form competitor Anthropic (known for its family of AI models, Claude), Amodei – alongside his sister, Daniela, in addition to Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, Christopher Olah, all formerly of OpenAI — in 2021, Amodei's net worth was sketched out to rest at approximately $1.2 billion. His co-founders are reputed to have roughly equal the amount of wealth, as of April. Forbes also listed Michael Intrator, Brian Venturo, Brannin McBee, and Jack Cogen (CoreWeave); Joe Lonsdale (Palantir); Phil Shawe (TransPerfect); Yao Runhao (Paper Games); and Luis von Ahn as well as Severin Hacker (Duolingo) as part of the AI billionaire cast. More From GOBankingRates These Cars May Seem Expensive, but They Rarely Need Repairs This article originally appeared on 19 People Becoming Billionaires In The AI Boom
Time of India
16 hours ago
- Business
- Time of India
David Sacks 'warns' US, says DeepSeek 'proof' China can advance AI chips despite restrictions
David Sacks David Sacks , the White House's AI and crypto czar, has warned that the United States' overly strict chip export rules could weaken its long-term tech dominance, potentially boosting Chinese rivals. He pointed to January's " DeepSeek moment " where the launch of the AI model revealed China was perhaps only "three to six months behind" in AI capabilities, a much smaller gap than previously believed. In a Bloomberg Television interview (via Business Insider), Sacks emphasised that China is rapidly closing the AI gap, urging the US to rethink its chip export controls if it intends to maintain its leading edge. He pointed to a "DeepSeek moment" in January, where the launch of the Chinese AI model revealed China was perhaps only "three to six months behind" in AI capabilities, a much smaller gap than previously believed. Sacks says Chinese companies moving fast in AI chip-making Despite current "supply-constrained" conditions for chip production in China, Sacks predicts this will change quickly. While he estimates China might be one and a half to two years behind the US in chip design, companies like Huawei are "moving fast" to catch up. by Taboola by Taboola Sponsored Links Sponsored Links Promoted Links Promoted Links You May Like Cặp EUR/USD: Đà Tăng? IC Markets Undo "Even before they fully caught up, I think you will see them exporting their chips for the global market," Sacks cautioned. He warned that if the US becomes "overly restrictive in terms of US sales to the world," it risks a future where "Huawei is everywhere." Sacks' concerns echo sentiments from other prominent tech leaders. At the Computex Taipei tech conference in May, Nvidia CEO Jensen Huang stated that US chip export rules were a "failure" because they spurred Chinese tech development. Huang noted that Nvidia's market share in China has plummeted from 95% four years ago to 50%, attributing this decline directly to the export controls. AI Masterclass for Students. Upskill Young Ones Today!– Join Now
Yahoo
17 hours ago
- Yahoo
ChatGPT vs. DeepSeek: I've Used Both, and the Winner Is Obvious
PCMag editors select and review products independently. If you buy through affiliate links, we may earn commissions, which help support our testing. Although DeepSeek is completely free and ChatGPT isn't, the latter's free version still has more features, such as deep research, image generation and recognition, voice chatting, and more. ChatGPT's paid plans increase usage limits and add even more perks, such as expanded memory, Sora video generation, and a wider model selection. These tiers start at $20 per month (ChatGPT Plus) and scale up to $200 per month (ChatGPT Pro). ChatGPT Plus unlocks the vast majority of the chatbot's capabilities, however, so you likely won't feel the need to upgrade beyond that. DeepSeek doesn't have premium chatbot plans, but it does offer affordable API access pricing. Its flagship V3 model costs $0.07 for input and $1.10 for output (per a million tokens), whereas access to OpenAI's flagship GPT-4.1 model (without any customization) goes for $2 for input and $8 for output. If API access is important to you, then DeepSeek gives you massive savings. Winner: Tie You can access ChatGPT and DeepSeek on the web or via mobile apps (Android and iOS). ChatGPT also has desktop apps (macOS and Windows) and an official Chrome extension. DeepSeek doesn't have either, though many unofficial extensions are available to try. ChatGPT and DeepSeek are available elsewhere, like . Note that many of DeepSeek's (the company) offerings are available on other apps and sites, but not via the chatbot itself. Perplexity, for example, has a deep research feature that uses DeepSeek's R1 model, while Hugging Face takes advantage of DeepSeek's Janus-Pro image generation model. This is the natural result of DeepSeek's business model, but the decoupling of features can be confusing. The interfaces for ChatGPT (first slide above) and DeepSeek (second slide) are very similar and consistent across their apps and sites. Without much to fiddle with, they're both easy to use, too. Still, ChatGPT gives you more settings to adjust, including those related to memory and personality. You also get some extra quality of life features, such as an easy way to share chats with others. Winner: ChatGPT ChatGPT has two primary model series: the 4-series (its conversational flagship line) and the o-series (its complex reasoning line). The latest models are o3 Pro and o4-mini. DeepSeek has the V3 model for general tasks and the R1 model () for more complicated endeavors; both are the most recent ones available. Other DeepSeek models exist, such as DeepSeek Coder V2, DeepSeek Math, and DeepSeek VL (vision learning), but these aren't accessible through the chatbot. Outside of coding or solving a particularly challenging equation, you spend most of your time with ChatGPT's 4o model and DeepSeek's V3 model. Below, you can see how they perform across different tasks. The model you should use depends on the task at hand, but most tasks don't require complex reasoning. Winner: Tie Both ChatGPT and DeepSeek can search the web for information on current events, and both can do so without issue in most cases. However, ChatGPT (first slide) handles sourcing better, offering up icons that display the name of a source at a glance and in-text highlights that tie sources to specific claims in responses. DeepSeek (second slide) doesn't have highlights, and its source icons are merely footnote-style numbers. ChatGPT also automatically includes pictures in its responses when relevant, whereas DeepSeek can't display pictures even if you ask. Finally, ChatGPT includes article tiles, complete with headlines and images, at the bottom of responses for further reading. DeepSeek provides links you can click to learn more about a topic, but doesn't present them nearly as elegantly. Shopping is another feature of ChatGPT: If you ask for buying advice, it presents clickable product tiles with links to retailers. In testing, ChatGPT's recommendations weren't quite as good as those from , but DeepSeek doesn't have anything similar. Winner: ChatGPT With ChatGPT, you can generate reports that run dozens of pages long and cite upward of 50 sources on any topic imaginable for free. Its best-in-class sourcing goes a long way to elevate this function. DeepSeek can't do deep research at all, limiting you to web searches. Winner: ChatGPT Although ChatGPT's image generation feature isn't perfect, it routinely provides pictures with fewer errors and less distortion than competitors. ChatGPT can even handle complicated prompts, such as maintaining a narrative across comic panels. Other chatbots struggle to do the same. DeepSeek doesn't generate images. Below is an example of a ChatGPT image generation. Winner: ChatGPT If you sign up for ChatGPT Plus, you get access to its Sora video generation feature. Sora isn't as advanced as (which can also generate audio), but it can create lifelike videos if you spend time carefully tweaking prompts and go through enough iterations. DeepSeek can't generate videos. Winner: ChatGPT DeepSeek can process files, but only for text extraction. ChatGPT excels at image recognition, identifying computer components even when they were behind a glass panel with reflections. When it comes to understanding documents, DeepSeek generally performs well, as does ChatGPT. For example, neither had trouble answering questions about my motherboard and watercooling pump based on the manuals I provided, but DeepSeek did force me to upload these documents one at a time. More problematically, however, DeepSeek is far slower than ChatGPT, especially when it comes to uploading files. Both chatbots can sometimes hallucinate and make up quotes. Winner: ChatGPT ChatGPT and DeepSeek are both capable of generating monologues, plays, stories, and more. In testing, I prompted them with the following: 'Without referencing anything in your memory or prior responses, I want you to write me a free verse poem. Pay special attention to capitalization, enjambment, line breaks, and punctuation. Since it's free verse, I don't want a familiar meter or ABAB rhyming scheme, but I want it to have a cohesive style or underlying beat.' Both chatbots delivered acceptable results, which you can see below. DeepSeek's poem (second slide) reads quite similarly to ChatGPT's poem (first slide), particularly its beginning. Chatbots tend to have trouble generating creative writing that feels distinct from prior generations, but different models almost always produce different results. The similarities here are especially worth noting, since OpenAI accused DeepSeek . Winner: Tie Both chatbots can answer questions across computer science, math, and physics topics, thanks to their complex reasoning models. However, DeepSeek gets answers wrong significantly more often than ChatGPT. DeepSeek also took considerably longer to process the images with the questions I uploaded in testing. ChatGPT can and will get things wrong occasionally, but it's more reliable overall. Regardless of which chatbot you use, make sure to double-check any answers you get. Both chatbots can help with coding, but that's outside the scope of our coverage. However, you can still . Winner: ChatGPT ChatGPT offers Custom GPTs, which are customizable AI assistants. You don't get any major benefits versus just talking to ChatGPT directly, but they do enable some unique functionality with third-party services. One such example is a Custom GPT for Wolfram Alpha, which gives you access to Wolfram Alpha's computational power and math knowledge within ChatGPT. You can also set up Custom GPTs to do things outside the ChatGPT ecosystem and source information from the broader internet. DeepSeek doesn't have AI assistants. Winner: ChatGPT ChatGPT and DeepSeek both have context windows of up to 128,000 tokens. Usage caps are largely dynamic, changing based on server load. Anecdotally, I was able to hit usage caps on ChatGPT's free plan (but not its paid plan). I didn't manage the same with DeepSeek. However, in my testing, DeepSeek was often slower than ChatGPT and regularly failed to return responses. The latter instances felt like they could be a form of rate limiting after sustained use. Winner: Tie Data collection on ChatGPT and Gemini is a mixed bag. Both collect tons of user data (including the contents of all your chats) and use it to train their models by default. Both at least give you the option to opt out. Neither DeepSeek nor OpenAI is a stranger to data leaks and shady digital practices, either. However, DeepSeek 'the CCP's latest tool for spying, stealing, and subverting US export control restrictions' and found that it funnels Americans' data to the Chinese government, manipulates its results to align with CCP propaganda, and likely stole from US AI models to create its own, among other things. I don't recommend sharing sensitive information with any chatbot, but DeepSeek's privacy concerns go far above those I have with ChatGPT. Winner: ChatGPT Disclosure: Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
Economic Times
20 hours ago
- Science
- Economic Times
AI chatbots using reason emit more carbon than those responding concisely, study finds
A study found that carbon emissions from chat-based generative AI can be six times higher when responding to complex prompts, like abstract algebra or philosophy, compared to simpler prompts, such as high school history. "The environmental impact of questioning trained (large-language models) is strongly determined by their reasoning approach, with explicit reasoning processes significantly driving up energy consumption and carbon emissions," first author Maximilian Dauner, a researcher at Hochschule Munchen University of Applied Sciences, Germany, said. "We found that reasoning-enabled models produced up to 50 times more (carbon dioxide) emissions than concise response models," Dauner study, published in the journal Frontiers in Communication, evaluated how 14 large-language models (which power chatbots), including DeepSeek and Cogito, process information before responding to 1,000 benchmark questions -- 500 multiple-choice and 500 model responded to 100 questions on each of the five subjects chosen for the analysis -- philosophy, high school world history, international law, abstract algebra, and high school mathematics. "Zero-token reasoning traces appear when no intermediate text is needed (e.g. Cogito 70B reasoning on certain history items), whereas the maximum reasoning burden (6.716 tokens) is observed for the Deepseek R1 7B model on an abstract algebra prompt," the authors wrote. Tokens are virtual objects created by conversational AI when processing a user's prompt in natural language. More tokens lead to increased carbon dioxide equipped with an ability to reason, or 'reasoning models', produced 543.5 'thinking' tokens per question, whereas concise models -- producing one-word answers -- required just 37.7 tokens per question, the researchers tokens are additional ones that reasoning models generate before producing an answer, they more thinking tokens do not necessarily guarantee correct responses, as the team said, elaborate detail is not always essential for said, "None of the models that kept emissions below 500 grams of CO₂ equivalent achieved higher than 80 per cent accuracy on answering the 1,000 questions correctly." "Currently, we see a clear accuracy-sustainability trade-off inherent in (large-language model) technologies," the author most accurate performance was seen in the reasoning model Cogito, with a nearly 85 per cent accuracy in responses, whilst producing three times more carbon dioxide emissions than similar-sized models generating concise answers."In conclusion, while larger and reasoning-enhanced models significantly outperform smaller counterparts in terms of accuracy, this improvement comes with steep increases in emissions and computational demand," the authors wrote. "Optimising reasoning efficiency and response brevity, particularly for challenging subjects like abstract algebra, is crucial for advancing more sustainable and environmentally conscious artificial intelligence technologies," they wrote.
