• AI Drop
  • Posts
  • ByteDance's Secret Use of OpenAI's Tech

ByteDance's Secret Use of OpenAI's Tech

Plus Fact Accuracy Assessment of ChatGPT with Bard, Claude and Copilot

Welcome back for the Latest AI Drops!

Today’s Drops:

  • ByteDance's Secret Use of OpenAI's Tech

  • ChatGPT,Bard, Claude and Copilot Fact Accuracy Assessment

  • Trending AI Startups “ GoLinks” & “Nunu.ai”

  • Trending on X “SceneWiz3D”

  • AI News: Deloitte utilizes AI for internal employee reallocation to prevent layoffs, while the 2024 U.S. presidential election increasingly incorporates AI in campaigning amidst ethical concerns. SenseTime's stock falls sharply following the death of its co-founder, amid a pivot to generative AI and new ventures in autonomous driving and healthcare.

  • Trending GitHub Projects

  • AI Tools

Read Time: 4.5 minutes

ByteDance's Secret Use of OpenAI's Tech

ByteDance, TikTok's parent company, has reportedly been using OpenAI's technology to develop its own large language model, potentially violating OpenAI's terms of service. According to The Verge, ByteDance has secretly utilized OpenAI's API for Project Seed, their AI project. Internal documents and communications suggest that ByteDance employees were instructed to use "data desensitization" to mask their actions. Despite these allegations, ByteDance claims it is licensed to use GPT APIs by Microsoft and uses its self-developed model for its chatbot, Doubao, in China.

OpenAI has confirmed that ByteDance's account was suspended while they investigated the matter. OpenAI spokesperson Niko Felix stated that ByteDance's use of the API was minimal, but they will terminate the account if necessary after their investigation.

This incident raises significant questions about ethical practices in the competitive AI industry and the implications of such actions on global tech relations, especially considering U.S. regulators' ongoing concerns over ByteDance's potential data-sharing with the Chinese government.

ChatGPT,Bard, Claude and Copilot Fact Accuracy Assessment

In a recent article by David Gewirtz, a unique experiment was conducted to test the accuracy of various AI tools in fact-checking. The author utilized ChatGPT to generate 150 facts about U.S. states, later cross-verifying these with other AI platforms such as Google's Bard, Anthropic's Claude, and Microsoft's Copilot.

The experiment began with ChatGPT producing facts and images for each state in collaboration with DALL-E 3. While the outcomes were predominantly accurate, there were notable errors, including geographical misplacements of landmarks. The subsequent phase involved assessing the factual accuracy through different AI models. Claude generally concurred with the facts, providing some nuanced clarifications. In contrast, Copilot faced challenges, often merely echoing the facts instead of scrutinizing them. On the other hand, Bard demonstrated superior performance in the fact-checking task, though it occasionally overcorrected or overlooked finer details.

This experiment not only showcases the varying capabilities of current AI technologies in handling detailed and complex tasks but also emphasizes the critical role of human oversight in fact verification. It serves as a reminder of the AI's evolving yet imperfect nature, especially in processing and verifying intricate data.

Trending AI Startups

GoLinks has introduced GoSearch, an AI-powered enterprise search engine that streamlines access to internal data, promising enhanced efficiency and privacy in handling company information.

Nuni.ai is building the first multimodal AI agents to help create better games.Check out the launch video

Trending on X

AI News

Deloitte is leveraging AI to reallocate employees to more promising roles within the firm, aiming to prevent layoffs. This strategy, highlighted in a Bloomberg report, involves using AI to assess and reassign employees based on their skills. The initiative is part of Deloitte's response to labor market changes influenced by AI advancements. It's an effort to balance the workforce amid rapid post-pandemic expansion and an uncertain economy, as the firm, like others in the consulting sector, faces the challenge of integrating a large number of new hires. Link 

The 2024 presidential election in the U.S. is witnessing an increased use of AI in campaigns and government operations. AI's role includes data analysis, message crafting, and social media strategy, potentially reducing the need for large campaign teams. However, this shift is challenged by public skepticism and a lack of regulation, especially concerning the use of deepfakes. AI offers the potential for more personalized and accessible political campaigns, but its successful and ethical integration depends on effective regulation and overcoming public distrust. Link

SenseTime Group Inc.'s stock fell significantly following the death of its co-founder, Tang Xiao’ou. This event caused an 18% decline in share value, the largest since July 2022. Tang was a prominent figure in China's AI sector and held a significant stake in the company. Despite facing challenges like U.S. sanctions and accusations of inflated revenues, SenseTime is refocusing on generative AI and new initiatives in autonomous driving and healthcare. Link

Welcome to the innovative world of GitHub projects!

openai/weak-to-strong: This project contains code for implementing weak-to-strong generalization in binary classification tasks, with a focus on fine-tuning pretrained language models and training against labels from another language model, and also includes code for weak-to-strong in vision models (AlexNet -> DINO on ImageNet)​​.

threestudio-project/threestudio: Threestudio is a unified framework for 3D content generation​​.

damo-vilab/i2vgen-xl: The official repository for VGen, a holistic video generation ecosystem built on diffusion models for video generation​​.

zulip/zulip: Zulip is an open-source team chat server and web application designed to help teams stay productive and focused​​.

microsoft/promptbench: Promptbench is a unified evaluation framework for large language models​​.

InterviewJARVIS is an AI-tool that simulates personalised interviews, offering over 200 question and answers based on your resume.

THERAi is a supercharged AI that evolves with you. Built on GPT-4, it offers tailored conversations and advanced tools like essay building, humanization, image generation and sentiment analysis.

Giftruly is an AI assistant for personalized gift ideas

Snoooz is an AI Out-of-Office (OOO) assistant that can send personalized OOO responses, automate OOO tasks, assign backups, create coverage plans, and so much more.

Captions for photos creates post captions with AI