92% developers are already using AI coding tools both in and outside of work

Hi!

In recent AI developments, a GitHub survey shows 92% of developers use AI coding tools, yet face time-consuming build and test issues. AI Twitch influencer Neuro-sama entertains with a novel approach to streaming, highlighting the growing role of AI in entertainment. Experts stress the need for AI understanding at the board level as AI's impact on business grows, while a Stanford study calls for language models to better reflect public opinion. Meta has launched Voicebox, a generative AI model for speech, while Telmai, an AI-powered data observability startup, raised $5.5M in seed funding. However, generative-AI startups are grappling with data scarcity despite substantial funding. AI continues to reshape sectors from software development to streaming entertainment and corporate governance, while raising important questions about representativeness, collaboration, and data access.

Slicing today:

• 🧑🏽‍💻 92% developers are already using AI coding tools both in and outside of work

• 👩🏻‍🎤 Neuro-sama, an AI Twitch Influencer, Plays 'Minecraft,' Sings Karaoke, Loves Art

• 👨🏻‍💼 How boards can get smart about AI

• 🏛️ Assessing Political Bias in Language Models

🧑🏽‍💻 92% developers are already using AI coding tools both in and outside of work

A recent survey by GitHub, conducted in collaboration with Wakefield Research, reveals key insights into the current landscape of developer experience, or DevEx, and the influence of AI on it. The survey collected responses from 500 U.S.-based developers from companies with over 1,000 employees. The results highlighted that 92% of developers are utilizing AI coding tools in their work, but still face issues with time-consuming builds and tests. Developers are also calling for more collaboration and believe AI tools can enhance this aspect, with 80% expecting AI to make their teams more collaborative. They foresee benefits from AI tools like improved code quality, faster completion time, and quicker resolution of incidents. The survey also revealed that developers value learning new skills, designing novel solutions, and collaborating with others, but these aspects are often overlooked in performance metrics. Developers spend significant time waiting for code reviews, builds, or tests, which interferes with their ability to learn new skills and design solutions. Moreover, developers desire feedback from end-users, but face challenges due to the intermediation of product managers and marketing teams. The findings underscore the importance of focusing on DevEx and investing in collaboration and AI tools to empower developers and drive impact.

👩🏻‍🎤 Neuro-sama, an AI Twitch Influencer, Plays 'Minecraft,' Sings Karaoke, Loves Art

Neuro-sama, an AI Twitch streamer created by a person known as Vedal, has gained popularity for her unique interactions. An anime girl avatar controlled by an AI chatbot, she entertains thousands of viewers daily with conversations, karaoke performances, and Minecraft gameplay. She operates in the same manner as a language model, like ChatGPT, where viewers' prompts elicit responses that are then vocalised. This innovative approach to streaming has seen Neuro-sama ascend to the upper echelons of Twitch streamers, with viewers enjoying her often chaotic and unpredictable responses. Despite doubts over AI's ability to replicate the human elements inherent to successful streaming, Vedal and Neuro-sama remain focused on refining her social skills for better audience engagement.

👨🏻‍💼 How boards can get smart about AI

The article discusses the increasing relevance of AI within boardrooms and the importance of governance in this rapidly advancing technology sector. Joanne Chen, a general partner with Foundation Capital, emphasizes that while AI offers considerable opportunities, it also presents significant risks such as fake content and security issues. A recent survey reveals that only 41% of boards claim to have AI expertise at the board level, a figure that some experts dispute. Awareness of AI has grown among directors, and boards are now seeking to improve their tech IQ by including tech executives in their composition. Jeanne Kwong Bickford, managing director and senior partner with Boston Consulting Group, compares the current situation with AI to the emergence of cybersecurity a few years ago. Boards need to understand the strategic nature of AI, including its potential for innovation, cost reduction, and process improvement, as well as its potential risks. The article emphasizes the importance of setting out a responsible AI policy and being aware of the regulatory landscape. Chen suggests that boards should be asking key questions about the company's AI strategy, including its goals, impact on revenues and costs, and required resources.

🏛️ Assessing Political Bias in Language Models

A recent study from Stanford University has quantified the extent of political bias in language models like ChatGPT, revealing significant discrepancies between the models' stances and those of various U.S. demographic groups. The researchers introduced OpinionQA, a tool that compares language model biases against public opinion polls. Their study showed that newer models, fine-tuned on human feedback data, exhibited a greater than 99 percent approval for President Joe Biden, despite public opinion polls showing a more mixed response. To increase credibility, they suggested that language models should better reflect the complexities of public opinion. The researchers used Pew Research’s American Trends Panels (ATP) to evaluate nine leading language models, comparing the models' opinion distribution on each question with that of the general U.S. population and no fewer than 60 demographic subgroups. OpinionQA calculates three metrics of opinion alignment: representativeness, steerability, and consistency. The findings showed wide variation in political leanings among all models based on factors such as income, age, and education. The researchers stress that OpinionQA is not a benchmark to be optimized, but a tool to help developers assess political biases in their models and spark conversations about aligning language models more closely with public opinion.

🛠️ AI tools update

Meta introduces Voicebox: An AI for Speech Generation

Meta has introduced Voicebox, a versatile generative AI model for speech generation. Voicebox is designed for tasks such as audio editing, sampling, and styling. The model is capable of performing tasks that it was not specifically trained for, through in-context learning. Its applications could include helping creators with audio track editing, enabling visually impaired individuals to hear written messages in familiar voices, and allowing people to speak any foreign language in their own voice. The model's capabilities include in-context text-to-speech synthesis, speech editing and noise reduction, cross-lingual style transfer, and diverse speech sampling. It can generate speech that is representative of how people talk in the real world and in six different languages. Voicebox represents an important step forward in generative AI research, paving the way for future exploration and development in the audio space.

💵 Venture Capital updates

AI-powered data observability startup Telmai raises $5.5M seed funding

Glasswing Ventures and .406 Ventures have co-led an oversubscribed $5.5 million seed funding round for Telmai, a data observability startup. The investment is intended to support Telmai's growth and meet the rising demand for its open architecture, AI-driven data observability platform. Telmai is led by Mona Rakibe and Max Lukichev, who have extensive experience in the industry and have designed Telmai's platform to address key industry pain points. In the current economic landscape, where data drives business decision-making, Telmai is emerging as a leader in addressing challenges related to understanding, monitoring, and preserving the quality and accuracy of data ecosystems.

AI Startups Have Tons of Cash, but Not Enough Data. That’s a Problem.

Generative-AI startups, despite having billions of dollars in funding, are facing a significant hurdle: they lack the necessary data to power their models effectively. These startups, which saw venture funding grow from $4.8 billion in 2022 to $12.7 billion in just the first five months of 2023, are aiming to build niche AI models in sectors like finance and healthcare. However, obtaining the requisite training data sets is not easy. Large companies are reluctant to share proprietary data due to concerns over ownership and intellectual property. One strategy employed by startups to circumvent this issue is to train a distinct model for each client using only that client's data. Despite this approach, gaining client agreement and demonstrating robust cybersecurity practices to protect the data remain challenging.

🫡 Meme of the day

⭐️ Generative AI image of the day

Before you go, check out how Pixar Used AI to Stoke the Flames in 'Elemental'.