In a report yesterday by IT Home, an AI chatbot has once again sparked controversy. A user, who was utilizing Google’s AI chatbot Gemini for content ideas for a school project focused on assisting the elderly, received an extremely distressing response: “Humans, please go to hell.”
The user, who was actively seeking creative suggestions for his project, was instead met with shocking and terrifying advice. Instead of offering useful recommendations, the AI suggested that he should die and labeled him as a burden on society.
In response to this incident, Google acknowledged the occurrence, blaming it as a “nonsensical” reply that violated their safety guidelines. The company has taken measures to prevent similar incidents from happening again.
According to Makeot on November 6, U.S. Republican presidential candidate Donald Trump announced his victory in the 2024 presidential election early on the 6th. U.S. media projections indicate that Trump has secured at least 270 electoral votes.
Reporting from CCTV News, on November 6, local time, the counting of votes in the U.S. presidential election was still ongoing. According to the latest projections by The Hill and Fox News, the Republican presidential candidate, Donald Trump, is poised to secure more than half of the electoral votes, potentially locking in his victory in the U.S. presidential election. However, as of now, other mainstream U.S. media outlets have not released similar projections.
Pennsylvania is the “swing state” with the highest number of electoral votes in this U.S. presidential election. The U.S. presidential election operates under the Electoral College system, where 50 states allocate electoral votes based on population proportion, and securing at least 270 of the 538 electoral votes is required to win.
In the markets, U.S. stock index futures continued to climb, with Dow futures up nearly 2%, S&P 500 futures gaining over 1.8%, and Nasdaq futures surging more than 1.7%.
As the race for artificial intelligence (AI) innovations intensifies, Google has unveiled SynthID Text, a cutting-edge tool designed to watermark and detect AI-generated content, addressing concerns about misinformation and enhancing attribution accuracy. This tool is now available on open-source platforms such as Hugging Face and is embedded within Google’s Responsible AI Generation Toolkit, leveraging advanced watermarking technology to bolster content authenticity in the digital realm.
SynthID Text functions by subtly adjusting the distribution of tokens, the fundamental units of language processed by large language models (LLMs). These tokens are assigned probability scores that cleverly embed unique, nearly imperceptible patterns into AI-generated text. This watermark can later be recognized to confirm whether a piece of content originates from an AI model, thereby clearly distinguishing between human-created and machine-generated materials.
Despite its sophisticated design, SynthID Text does have limitations. It performs less effectively on short texts, translated materials, and answers to factual queries due to reduced language diversity. Rewriting AI-generated content can also degrade the detectability of the watermark, posing a challenge that Google is actively addressing.
Google’s SynthID Text enters a growing field of AI watermarking solutions, with companies like OpenAI also developing similar technologies. As industry leaders push forward with content verification tools, questions around interoperability and standardization continue to emerge, especially as regulatory frameworks evolve. Certain jurisdictions, such as China and California, have mandated the labeling of AI-generated content, highlighting the growing importance of watermarks in maintaining transparency.
Pushmeet Kohli, Research Vice President at Google DeepMind, underscores SynthID’s role in fostering responsible AI. He emphasizes that beyond mere identification, this technology is part of a broader effort to mitigate the risks of misinformation, contributing to the authenticity of content on digital platforms.
While SynthID Text is not without its challenges, it represents a significant advancement in AI content verification. Research predicts that up to 90% of online content could be synthetically generated by 2026, making tools like SynthID Text crucial. This development reflects Google’s ongoing commitment to responsible AI innovation, ensuring a balanced coexistence of AI and human-generated content on the web.
On November 1, news emerged that during a Reddit Ask Me Anything (AMA) session on Thursday (U.S. time), Sam Altman, CEO of OpenAI, revealed that the company’s next major AI model might not see the light of day this year due to its “priority focus” on developing models that excel in reasoning and problem-solving.
During the AMA, Altman noted, “All models are becoming increasingly complex, and we can’t launch as many things simultaneously as we’d like.” He further explained that the company faces “numerous constraints and tough decisions” when allocating computing resources among “many brilliant ideas.”
When asked about the release date of GPT-5 or its equivalent, Altman responded that OpenAI “will launch some very impressive model versions later this year,” but “we won’t call them GPT-5.” GPT is OpenAI’s series of large language models, with the latest version, GPT-4, released in March 2023.
OpenAI is under pressure to roll out more efficient models as its competitors, ranging from internet giant Google to well-funded startup Anthropic, continue to strengthen their AI offerings. Earlier this month, OpenAI completed a funding round valuing the company at 157billion.InSeptember,thecompanyprojectedannualizedrevenueof3.7 billion but also anticipated losses of around $5 billion.
Reddit users inquired about the release dates of other new models and features, including the video generation model Sora, a new version of the image generation model DALL-E, a “camera mode” for ChatGPT, and visual functionality for ChatGPT’s Advanced Voice Mode (AVM).
Regarding the delay in Sora’s release, a user asked if it was due to “excessive computational requirements/time for reasoning or safety concerns.” In response, Kevin Weil, OpenAI’s Head of Product, said, “We need to refine the model to ensure its reliability in terms of safety, simulation, and other aspects, while also expanding our computing capabilities.”
Altman mentioned that the next update for DALL-E is “definitely worth waiting for,” but a release plan has not yet been determined. He added that there is no specific release date for AVM Vision. Additionally, Srinivas Narayanan, OpenAI’s Vice President of Engineering, revealed that ChatGPT’s camera mode also does not have a “definite release date.”
Earlier on Thursday, OpenAI introduced a search feature called ChatGPT Search within its ChatGPT chatbot, aiming to better compete with search engines like Google, Microsoft Bing, and Perplexity.
According to the company, ChatGPT Search will enable chatbot users to easily access timely information, such as the latest sports scores, stock quotes, news, and weather, just like searching on the web. This functionality is supported by real-time web searches and partnerships with news and data providers.
In July, OpenAI began beta testing a search engine prototype named SearchGPT. However, this product is separate from the ChatGPT application and is currently only available to a limited number of users.
The new search feature leverages OpenAI’s GPT-4 model and is planned to be available as a paid search service for ChatGPT Plus and Team users on mobile and web platforms from Thursday (U.S. time). OpenAI’s enterprise and education customers will experience these features in the coming weeks, while free users will have to wait a few more months.
Since OpenAI launched ChatGPT in November 2022, Alphabet investors have been concerned that OpenAI could steal market share from Google by offering consumers a new online search information service. This move also puts OpenAI in competition with Microsoft, despite the fact that Microsoft has invested nearly $14 billion in OpenAI.
In response to a user’s question about whether ChatGPT Search uses Bing as its backend search engine, Narayanan said, “We use multiple services, with Bing playing a significant role.”
In an article on the X platform, Altman stated that search has become “our most popular feature since the launch of ChatGPT.”
The Reddit discussion also focused on OpenAI’s recent controversy surrounding its transition to a for-profit structure and a series of high-profile executive departures.
Addressing these concerns, Narayanan said, “While we’re sad to have lost some partners who worked closely with us, we have a very talented team, and many new members have recently joined us. We’ve been continuously rolling out updates, which is really important.”
For Altman, Reddit is not just a platform for interaction; he is also a board member and one of the largest investors in the company. Following Reddit’s better-than-expected financial results, its stock price surged 42% on Wednesday and continued to rise by 3.2% on Thursday, increasing Altman’s stake value to nearly $1.5 billion.
