Ilya is back 👀

Ilya Sutskever, co-founder and former chief scientist of OpenAI, has announced the launch of a new AI startup called Safe Superintelligence Inc. (SSI). The company, which will focus on creating a safe and powerful AI system, aims to prioritize safety over commercial pressures. SSI’s approach integrates safety and capabilities, allowing it to advance its AI technology while maintaining a strong emphasis on safety. Sutskever’s announcement highlights the company’s unique business model, which insulates its operations from short-term commercial influences, enabling it to scale without being distracted by management overhead or product cycles.

SSI is co-founded by Daniel Gross, a former AI lead at Apple, and Daniel Levy, who previously worked at OpenAI. The company’s formation follows a period of significant changes at OpenAI, including Sutskever’s departure in May and subsequent hints at a new venture. Other notable resignations from OpenAI, such as AI researcher Jan Leike and policy researcher Gretchen Krueger, have cited concerns over safety processes being overshadowed by product development priorities. These departures underscore a broader industry tension between rapid AI advancements and the necessity for rigorous safety measures.

In contrast to OpenAI’s active partnerships with tech giants like Apple and Microsoft, SSI plans to concentrate solely on developing safe superintelligence as its first and only product for the foreseeable future. Sutskever’s commitment to safety-driven AI development reflects a strategic shift aimed at addressing the critical need for secure and ethical AI technologies. During an interview with Bloomberg, Sutskever emphasized that SSI will remain singularly focused on its mission to ensure AI safety, distancing itself from the typical commercial pressures faced by other major AI firms.

Automate your SEO with OTTO

OTTO SEO is the world’s first automated SEO AI agent. With OTTO, you can complete thousands of webpage optimizations in just a few clicks. Eliminate the manual work of SEO and save hundreds of hours for your marketing teams. OTTO can accomplish more for your website in 10 seconds than an entire team of strategists, content writers, and web developers can do manually in a month.

Simply add the OTTO pixel to your website header and start reviewing and deploying optimizations in the dashboard.

OTTO can resolve technical issues, deploy content optimizations, secure backlinks, and even publish new content directly to your website. You’ll never have to leave the dashboard to complete thousands of SEO tasks. Once you approve optimizations in the dashboard, OTTO will make those changes to your website. Instantly. No matter your CMS.

Call Centers Introduce ‘Emotion Canceling’ AI as a ‘Mental Shield’ for Workers

Call center jobs are notoriously stressful, with employees often facing disgruntled customers and responding with scripted replies. In an effort to mitigate the emotional toll, some companies are introducing AI systems as a form of emotional support. SoftBank has developed a voice-altering technology called SoftVoice, which detects hostile tones in customer voices and alters them to sound calm, providing a "mental shield" for employees. By 2025, SoftBank plans to sell this technology more broadly.

First Horizon Bank also explored using AI to support their call center employees. The bank considered implementing a system that detected when an employee was stressed and sent them a relaxing video montage featuring personal photos and calming music. Initially, this approach seemed promising, with reported reductions in burnout levels, but the bank ultimately decided not to adopt the system widely.

These AI initiatives, while aiming to support employees, also hint at a future where AI could replace call center operators entirely. By honing their ability to recognize and respond to emotional cues, these systems are gradually preparing to handle customer interactions autonomously. Until then, AI serves as a temporary solution to the high levels of stress and burnout that call center employees face daily.

China’s DeepSeek Coder becomes first open-source coding model to beat GPT-4 Turbo

Chinese AI startup DeepSeek has launched DeepSeek Coder V2, an open-source mixture of experts (MoE) code language model that surpasses state-of-the-art closed-source models such as GPT-4 Turbo, Claude 3 Opus, and Gemini 1.5 Pro. Built on DeepSeek-V2, this model supports over 300 programming languages and excels in coding and math tasks, outperforming competitors on benchmarks like MBPP+, HumanEval, and Aider. This achievement marks the first time an open-source model has outperformed leading closed-source models, positioning DeepSeek Coder V2 ahead of Llama 3-70B and other models in its category.

DeepSeek Coder V2, with 16B and 236B parameter options, enhances the capabilities of its predecessor by increasing language support to 338 and extending the context window to 128K. The model achieved impressive scores on benchmarks evaluating code generation, editing, and problem-solving capabilities, such as 76.2 on MBPP+, 90.2 on HumanEval, and 73.7 on Aider. Additionally, it demonstrated strong mathematical capabilities on MATH and GSM8K benchmarks. The model's performance is attributed to its foundation on DeepSeek V2 and additional pre-training on a dataset of 6 trillion tokens, primarily code and math-related data.

In general reasoning and language understanding tasks, DeepSeek Coder V2 also performs admirably, scoring 79.2 on the MMLU benchmark. This score is significantly higher than other code-specific models and comparable to Llama-3 70B. While GPT-4o and Claude 3 Opus lead in the MMLU category, the development of DeepSeek Coder V2 indicates that open coding-specific models are closing the gap with state-of-the-art closed-source models. The model is available under an MIT license for research and commercial use, with options for download via Hugging Face or API access through a pay-as-you-go model.

Other stuff

• I Will F…ing Piledrive You If You Mention AI Again

• Not all ‘open source’ AI models are actually open: here’s a ranking

• France leads the pack for Generative AI funding in Europe, London has 3X the number of GenAI startups

• AI Training Data, In-Depth. From Diverse Inputs to Ethical Sourcing and Oversight

• Nvidia releases free LLMs that match GPT-4 in some benchmarks

• I Am Laura Kipnis-Bot, and I Will Make Reading Sexy and Tragic Again

• How to Fix “AI’s Original Sin”

• What happened when 20 comedians got AI to write their routines

Hope you enjoyed today's newsletter

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

OR