ChatGPT's Live Video (Vision) feature getting ready for a wider beta rollout

OpenAI appears to be gearing up for a broader beta rollout of its ChatGPT Live Video (Vision) feature, initially teased during the GPT-4o announcement in May 2024. This feature builds on the Advanced Voice Mode, which introduced a conversational element to ChatGPT, allowing users to engage naturally with the AI. The new Vision capability enhances this experience by enabling ChatGPT to interact with real-world visuals. During the announcement demo, the AI impressed by recognizing objects, associating them intelligently (like identifying a ball with a dog), and requiring minimal user input to respond effectively.

Recent updates in the beta version of ChatGPT (v1.2024.317) hint at the feature being named "Live Camera" when it moves to beta testing. Strings found in the code describe how users will tap a camera icon to let ChatGPT view and discuss their surroundings. However, the feature comes with a cautionary note advising against using it for live navigation or decisions that could affect health or safety.

Alpha testers have reported that the Vision feature works seamlessly and adds significant value to user interactions, offering practical and intuitive use cases. While OpenAI has not officially announced a release date, the presence of these strings strongly suggests the feature is nearing readiness. It’s expected to be available to ChatGPT Plus and paid subscribers, promising a major step forward in AI’s ability to integrate voice, vision, and natural conversation.

Also: ChatGPT 4o "Creative Writing" upgrade today is INSANE!

Capture Every Word With VoiceHub

Meet VoiceHub, the new productivity platform from Rev. It’s revolutionizing how businesses—from newsrooms to law firms to consultancies—handle their most valuable asset: conversations. Think of it as your team's AI-powered conversation hub.

While basic transcription tools might capture words, VoiceHub captures insights so your team can get everything they need from their conversations.

What sets VoiceHub apart? Their AI accuracy beats Microsoft, Google, and other leading providers in enterprise environments. But it's not just about accuracy—it's about changing the way you use those accurate transcripts. With VoiceHub, you get:

• Universal capture of audio and video across mobile, desktop, and meetings

• Best-in-class AI transcription in seconds

• Custom AI templates that automatically extract insights, action items, and key themes

• Enterprise-grade security with SOC 2 Type II compliance and SSO

• Seamless integration with major tools like Zoom and Slack

A Chinese lab has released a ‘reasoning’ AI model to rival OpenAI’s o1

DeepSeek, a Chinese AI research company funded by quantitative traders, has introduced DeepSeek-R1, a reasoning AI model designed to compete with OpenAI's o1. The firm claims that this new model, officially named DeepSeek-R1-Lite-Preview, matches the performance of OpenAI’s o1-preview on prominent AI benchmarks like AIME and MATH. Unlike traditional AI models, reasoning models like DeepSeek-R1 dedicate more time to processing queries, enabling them to better fact-check and avoid common pitfalls. However, this deliberate "thinking" process means that the model may take several seconds to deliver an answer, especially for complex queries.

Despite its achievements, DeepSeek-R1 has its shortcomings. Critics have pointed out the model's struggles with basic logic problems like tic-tac-toe, a challenge shared by its rival, o1. Additionally, DeepSeek-R1 has demonstrated vulnerabilities to "jailbreaking," where users manipulate the model to bypass safety measures, with one example involving the generation of a drug recipe. The model also restricts politically sensitive topics, likely in compliance with Chinese government regulations that require AI models to adhere to "core socialist values" and avoid controversial issues such as discussions about Xi Jinping or Tiananmen Square.

The development of reasoning models like DeepSeek-R1 comes at a time when traditional scaling laws—principles suggesting that increasing data and computing power will consistently improve AI models—are being questioned. As advancements in AI slow, companies are exploring new methodologies like test-time compute, which allows models to allocate additional processing time during inference to enhance performance. This approach has been hailed as a transformative shift in AI development, with Microsoft CEO Satya Nadella recently highlighting its significance during a keynote at Ignite 2024.

DeepSeek’s efforts are backed by High-Flyer Capital Management, a quantitative hedge fund leveraging AI for trading strategies. Known for its ambition to develop "superintelligent" AI, High-Flyer has invested heavily in infrastructure, reportedly building server clusters with 10,000 Nvidia A100 GPUs. By open-sourcing DeepSeek-R1 and introducing an API, the company aims to disrupt the AI landscape further, echoing the impact of its earlier DeepSeek-V2 model, which compelled competitors like Baidu and ByteDance to cut prices and release free versions of their models.

OpenAI releases a teacher’s guide to ChatGPT

OpenAI has introduced a new initiative aimed at integrating AI into classrooms, launching a free online course for K-12 teachers in collaboration with Common Sense Media. This one-hour program, divided into nine modules, covers foundational AI concepts and pedagogical applications of tools like ChatGPT. Already tested in districts like Agua Fria in Arizona and San Bernardino in California, OpenAI claims that 98% of participants found the course beneficial. Robbie Torney, Common Sense Media’s senior director, emphasized the importance of preparing educators for the transformative impact of AI in education.

However, the initiative has not been met with universal approval. Critics like Lance Warwick, a lecturer at the University of Illinois Urbana-Champaign, argue that the course's guidance on privacy and ethical use of AI is contradictory. While it advises teachers not to input sensitive student data, it also provides examples of activities that seem to require such data. Concerns over the potential misuse of AI-generated content and inherent biases in generative AI have further fueled skepticism. Sin à Tes Souhaits, an educator and visual artist, worries about OpenAI’s intentions regarding the intellectual property generated using its tools, pointing out a lack of clarity on how such content might be stored or reused.

The broader debate over AI in education persists. UNESCO has called for stricter regulation of AI in schools, including age limits and data privacy safeguards, but progress has been slow. Critics like Tes Souhaits also argue that OpenAI’s course promotes its own tools exclusively, which may contribute to monopolistic tendencies in the AI industry. Despite these concerns, some educators, like Josh Prieur of Prodigy Education, see potential in AI to streamline educational tasks if implemented thoughtfully, emphasizing the need for safeguards to protect both students and teachers.

The education market remains a significant focus for OpenAI, which has been actively pursuing partnerships and developing specialized products like ChatGPT Edu for universities. While some studies suggest mixed impacts of AI tools on learning outcomes, OpenAI’s vision is to augment, not replace, traditional teaching methods. Yet, with many educators still skeptical about AI’s role in classrooms, adoption may hinge on addressing ethical, practical, and pedagogical concerns more robustly.

Other stuff

• Oops! OpenAI just deleted important legal data in a lawsuit from The New York Times

• OpenAI's assault on Google continues. The ChatGPT owner recently considered launching a browser

• DeepSeek, A Chinese lab has released a ‘reasoning’ AI model to rival OpenAI’s o1

• The AI Reporter That Took My Old Job Just Got Fired

• Can Google Scholar survive the AI revolution?

• How Meta Uses LLMs to Improve Incident Response

• Prompt Injecting Your Way To Shell

• YouTube Shorts’ Dream Screen feature can now generate AI video backgrounds

• How Satya Nadella put Microsoft back on top

• AI Can ‘Hear’ When a Lithium Battery Is About to Catch Fire

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.