2025-01-29
OpenAI Furious DeepSeek Might Have Stolen All the Data OpenAI Stole from Us
OpenAI, Microsoft, and AI expert David Sacks allege that DeepSeek unfairly used OpenAI's data to develop a superior language model. Investigations by Bloomberg and Financial Times are examining if DeepSeek improperly trained its R1 model using OpenAI outputs, potentially through a technique called "distillation," which extracts knowledge from existing models. The situation is notable as OpenAI, often criticized for using data without authorization, is now raising concerns about similar practices by others.
Reactions
OpenAI is investigating allegations that DeepSeek may have improperly used data to train its R1 model. Claims of OpenAI being "furious" are considered unfounded and are seen as clickbait. Hacker News emphasizes the importance of accurate reporting and discourages misleading headlines.
I still like Sublime Text
Sublime Text remains a preferred choice in 2025 due to its speed, cross-platform compatibility, and features like Language Server Protocol (LSP), snippets, and multiple cursors. The software is considered "done," meaning it is stable and reliable, with the latest update released in January 2025, maintaining its relevance and efficiency. Sublime Text has significantly influenced modern editors, and its simplicity and distraction-free environment continue to attract users seeking a fast and efficient coding experience.
Reactions
Sublime Text is favored for its speed, simplicity, and lightweight nature, making it ideal for quick edits and small-scale coding tasks. While some users have transitioned to IDEs like VS Code or Zed for advanced features such as remote development and AI integration, Sublime Text remains popular for its stability and efficiency. The tool is often used alongside more feature-rich IDEs, highlighting its role as a complementary tool in a developer's toolkit.
Science YouTuber physicsgirl (Dianna Cowern) stands for the first time in 2 yrs
Reactions
Science YouTuber Physics Girl, Dianna Cowern, stood for the first time in two years after battling long COVID, sparking discussions on recovery experiences. The community shared diverse treatments and lifestyle changes, including dietary adjustments, exercise, and supplements, highlighting personal recovery journeys. Conversations also addressed challenges in obtaining adequate medical care and the frustration of being dismissed by healthcare professionals, fostering hope and encouragement for others facing similar health issues.
Why DeepSeek had to be open source
DeepSeek, a Chinese AI lab, has open-sourced its reasoning model R1, which is comparable to OpenAI's o1, to build trust in Western markets and address skepticism about Chinese tech, particularly regarding data security. The open-sourcing of R1 is also a strategic move to innovate efficiently amid export controls on advanced chips, highlighting the cost advantage of open-source models over proprietary ones like OpenAI. Open-source models, such as DeepSeek's R1, offer customizability and lower costs, challenging proprietary models, although OpenAI maintains a leadership position due to its first-to-market advantage and resources.
Reactions
DeepSeek's open-source AI model has initiated discussions on AI data practices, particularly following OpenAI's contentious use of YouTube data. DeepSeek provides a free alternative to OpenAI's expensive models, prompting a shift among users, though its full training process remains undisclosed, questioning the definition of "open source" in AI. The debate underscores the ongoing tension between proprietary and open-source AI models, with DeepSeek challenging existing norms and making AI more accessible, yet the future balance between these models remains uncertain.
I do not want AI to "polish" me
Reactions
The author criticizes AI writing tools for making text overly formal and impersonal, comparing them to the unpopular Microsoft Clippy. While AI tools can aid non-native speakers, there is concern they result in bland and uniform communication, stripping away individuality. The push for AI features in the tech industry is perceived as market-driven rather than based on actual user needs, with a preference for authentic human communication.
Questions censored by DeepSeek
DeepSeek-R1, an open-source model from China, is under scrutiny for being subject to Chinese Communist Party (CCP) censorship, with concerns about its refusal rate on sensitive topics. A dataset of 1,360 prompts was used to test the model, revealing that approximately 85% of prompts on topics like Taiwanese independence were censored, yet jailbreaking attempts showed the censorship could be easily bypassed. Future evaluations will compare the handling of sensitive topics between Chinese and American models, highlighting differences in censorship and content moderation.
Reactions
DeepSeek, a Chinese AI model, censors topics sensitive to the Chinese government, such as the Tiananmen Square Massacre, by deleting content triggered by specific keywords. Hosted versions of DeepSeek exhibit more censorship compared to local versions, sparking discussions on AI censorship and comparisons to Western models like ChatGPT. The situation underscores the broader issue of AI models being shaped by the cultural and political contexts of their development.
Goodbye, Slopify
A user expresses dissatisfaction with Spotify's Discover Weekly playlist, which now includes AI-generated music and covers. The user criticizes Spotify's app for being bloated and clunky, leading to their decision to unsubscribe. This situation highlights a growing concern about AI-generated content in music streaming services and its impact on user experience and support for real artists.
Reactions
Spotify's focus on algorithm-driven playlist customization and AI-generated music has led to user dissatisfaction, prompting some to leave the platform. Users have expressed frustration over issues such as incorrect metadata, unwanted podcast promotions, and missing features like profile swapping on family plans. As a result, many users are exploring alternatives like Apple Music, Tidal, and Qobuz, which offer better artist support and less reliance on algorithms.
Bacteria (and Their Metabolites) and Depression
Reactions
There is a potential link between bacteria, their metabolites, and depression, with some individuals reporting mood improvements after fasting, possibly due to microbiome changes. The discussion highlights the impact of dietary changes, such as removing wheat and sugar, on mental health, and the role of gut bacteria, genetics, and lifestyle in overall well-being. There is growing interest in understanding the connection between gut health and mood, as well as exploring personal genomics for enhanced health management.
DeepSeek's multi-head latent attention and other KV cache tricks
Key-Value (KV) caches enhance the speed of language models, such as ChatGPT, by reducing redundant computations, thus improving inference efficiency. KV caching decreases computational complexity from 𝑂(𝑛³) to 𝑂(𝑛²) by storing key and value information for each token, though it increases memory usage. Innovations in token selection, compression, and architectural redesigns, such as Multi-head Latent Attention, address the memory demands of KV caching, optimizing Transformer-based models.
Reactions
DeepSeek's article on multi-head latent attention and KV (Key-Value) cache techniques generated a lively discussion, highlighting both criticism and appreciation for AI-generated content. The article, co-authored with AI, aimed to offer a concise summary of complex research, utilizing PySpur for its visual appeal and loop support, with suggestions for further enhancements. The discussion emphasized the importance of transparency in AI-generated content and the potential for AI to improve writing efficiency.
Parkinsons patient "feels cured" with new adaptive deep brain stimulation device
Kevin Hill, a Parkinson's patient, reports feeling "cured" after receiving a groundbreaking brain implant that sends electrical signals to his brain, significantly alleviating his symptoms. The adaptive deep brain stimulation system, developed by Medtronic, automatically adjusts treatment based on brain activity, offering potential improvements in patients' quality of life. With approximately 153,000 people in the UK affected by Parkinson's, this innovation could have a substantial impact as the number of cases is expected to increase.
Reactions
A new adaptive deep brain stimulation (DBS) device is reported to significantly alleviate Parkinson's symptoms, with one patient claiming to "feel cured," though skepticism persists. Critics caution that such reports may offer false hope, as DBS treatments are not universally accessible or affordable, and they do not address all symptoms like dementia or psychosis. While DBS has been used for years and shows promise, more clinical trials are necessary to validate its effectiveness, and it is important to note that it is not a cure for Parkinson's disease.
Seagate: 'new' hard drives used for tens of thousands of hours
German customers have reported purchasing "new" Seagate hard drives that were actually used, with some showing extensive usage hours. The issue was identified by Heise.de after multiple reports concerning Seagate Exos HDDs, which were sold as new but lacked the standard five-year warranty. Seagate is investigating potential fraud, as the drives' SMART (Self-Monitoring, Analysis, and Reporting Technology) stats were reset, and the issue involves several retailers, including Amazon and Mindfactory.
Reactions
Seagate hard drives marketed as "new" in Germany were discovered to have been previously used for extensive periods, with their SMART data reset to conceal prior usage. The issue appears to involve multiple retailers, indicating a potential distributor error or fraudulent activity. Consumers are advised to verify drive usage using smartctl's FARM logs, though this requires updated software versions not universally available, raising concerns about purchasing hard drives from certain retailers, particularly on platforms like Amazon.
Discovery Coding
Discovery Coding is a programming method where coders write code to understand a problem, similar to discovery writing in fiction, where the story develops during the writing process. This approach may seem less organized compared to structured planning but can still yield rigorous and effective end-products by focusing on understanding system constraints. The programming community is encouraged to recognize and accept diverse approaches to software design, acknowledging that there is no single correct method, as it is a human process that varies among individuals.
Reactions
Discovery coding is a method where programmers write code to explore and understand a problem, rather than depending only on design documents. This approach can lead to more accurate time estimates and solutions by providing a hands-on understanding of the problem space. While it may result in shipping incomplete solutions, it is also seen as fostering innovation and adaptability, particularly in dynamic environments.
Apple and SpaceX link up to support Starlink satellite network on iPhones
Reactions
Apple is partnering with T-Mobile to support the Starlink satellite network on iPhones, aiming to improve connectivity in remote areas. This collaboration reflects a broader trend of integrating satellite technology into smartphones, alongside advancements in artificial intelligence (AI). Some users are concerned about Apple's association with Elon Musk, while others view it as a strategic move to enhance service offerings.
Libraries and Well-Being: A Case Study from The New York Public Library
A study by the New York Public Library reveals that libraries greatly improve well-being and community connection, particularly in lower-income areas. Key findings show that 92% of users feel calm after visiting, 74% feel better equipped to handle life, and 90% enjoy learning more. Libraries are identified as places fostering optimism, empathy, and community, serving as a powerful antidote to modern stressors.
Reactions
U.S. libraries are experiencing a shift as university systems reduce physical collections, opting for digital access due to decreased demand for physical books. Public libraries face additional challenges, often doubling as shelters for the homeless, which can impact their primary role of serving the community. Despite these issues, libraries continue to adapt by incorporating modern amenities like maker spaces, highlighting the ongoing debate on balancing traditional functions with changing societal needs.
Our phones are killing our ability to feel sexy (2024)
Reactions
The article discusses how smartphones may reduce our sense of sexiness by promoting constant digital immersion, which detracts from real-life social interactions and spontaneity. It argues that smartphones foster a lifestyle focused on instant gratification, leading to a decline in vibrancy and personal accountability for phone usage. The piece also explores broader societal issues, including the impact of technology on personal responsibility and the complexities of modern life.