Hiya, folks, welcome to TechCrunch’s regular AI newsletter. If you want this in your inbox every Wednesday, sign up here.

Two AI Startups Raise Nearly Half a Billion Dollars in Funding

This week, two startups developing tools to generate and suggest code, Magic and Codeium, raised nearly half a billion dollars combined. The funding rounds are significant, even by the high standards of the AI sector.

Magic, a generative AI coding startup, landed $320 million in investment from Eric Schmidt, Atlassian, and others. Notably, Magic has not yet launched a product or generated revenue.

Codeium, a competitor to GitHub Copilot, raised $150 million at a valuation of $1.25 billion. The funding rounds demonstrate the growing interest in generative AI startups, with VCs pouring billions into these companies.

Why Investors are Bullish on Coding

Investor enthusiasm for coding is no surprise. The reality is that coding isn’t a simple or affordable endeavour. Despite these challenges, there’s a growing demand from both companies and individual developers for solutions to simplify the more laborious aspects of coding.

Demand is driven by the need to streamline processes, making it easier and more efficient to develop software. This trend is expected to continue as technology advances and the importance of coding in various industries grows.

The Burden of Code Maintenance: A Dev’s Perspective

A recent survey reveals that the average developer spends approximately 20% of their workweek maintaining existing code, rather than focusing on writing new code. This staggering statistic highlights the significant amount of time and effort devoted to upkeep and maintenance.

Furthermore, a separate study found that excessive code maintenance, including addressing technical debt and fixing poorly performing code, costs companies a whopping $85 billion per year in lost opportunities. This striking figure underscores the importance of efficient code management and the need for developers to prioritize their work accordingly.

As developers strive to stay ahead of the curve, it’s essential to recognize the impact that code maintenance can have on productivity and profitability. By adopting strategies to streamline maintenance tasks and reduce technical debt, companies can minimize losses and redirect resources towards innovation and growth.

[Survey 1: https://stripe.com/reports/developer-coefficient-2018] [Study 2: https://adtmag.com/articles/2018/09/10/developer-survey.aspx]

Unlocking Developer Productivity with Generative AI

AI tools can significantly assist developers and firms, as many experts and consultants agree. According to a 2023 report by McKinsey, generative AI coding tools have the potential to enable developers to write new code in half the time and optimize existing code in roughly two-thirds of the original time.

McKinsey report

The Limitations of Coding AIs

While coding AIs have shown great promise in automating repetitive and mundane tasks, they are not a silver bullet for all development workloads. The McKinsey report highlights that certain complex tasks, such as those requiring familiarity with specific programming frameworks, do not necessarily benefit from AI.

In some cases, junior developers may even take longer to complete certain tasks when using AI compared to without it. This suggests that while AI can be a valuable tool for simplifying development workflows, it is not a replacement for human expertise and judgment in all situations.

The Limitations of AI in Code Development

Participant feedback reveals that developers are able to achieve high-quality code by actively iterating with AI tools. This suggests that AI is best used to augment developers, rather than replace them.

According to the co-authors, AI is no substitute for experience and understanding what makes up quality code. To maintain code quality, developers need to understand the attributes of good code and prompt the tool for the right outputs.

AI Coding Tools: Unresolved Security and IP Concerns

The widespread adoption of AI coding tools has raised concerns about their impact on software development. While these tools are touted for their ability to speed up the coding process, they also have unresolved security- and IP-related issues.

Some analyses suggest that these tools have led to an increase in more mistaken code being pushed to codebases over the past few years. This is a worrying trend, as it can compromise the quality and reliability of software systems.

Furthermore, AI coding tools trained on copyrighted code have been caught regurgitating that code when prompted in a certain way. This poses a significant liability risk to developers who use these tools, as they may inadvertently infringe on the intellectual property rights of others.

But that’s not dampening enthusiasm for coding AI from devs — or their employers, for that matter.

The Rise of Artificial Intelligence in Software Development

According to a 2024 GitHub poll, an overwhelming majority (upward of 97%) of developers have adopted AI tools in some form. The poll also revealed that between 59% to 88% of companies are now encouraging or allowing the use of assistive programming tools.

This growing trend towards AI adoption is likely driven by the numerous benefits it offers, including increased productivity and efficiency, improved code quality, and enhanced collaboration among team members. As AI continues to evolve and mature, it will be exciting to see how developers leverage these technologies to drive innovation and shape the future of software development.

The AI coding tools market is expected to reach a staggering $27 billion by 2032, according to Polaris Research. This projection is not entirely unexpected, especially considering Gartner’s prediction that by 2028, a whopping 75% of enterprise software engineers will use AI code assistants.

[Reference: https://www.polarismarketresearch.com/press-releases/ai-code-tools-market] [Reference: https://www.gartner.com/en/newsroom/press-releases/2024-04-11-gartner-says-75-percent-of-enterprise-software-engineers-will-use-ai-code-assistants-by-2028]

The Rise of Generative AI Coding Startups

The market is already hot, with generative AI coding startups like Cognition, Poolside, and Anysphere closing massive rounds in the past year. These startups have managed to convince investors and customers that their productivity gains outweigh their flaws.

Cognition, for instance, has raised $175 million at a $2 billion valuation just one month after its Series A round. Similarly, Poolside is raising $400 million at a $2 billion valuation for its supercharged coding copilot. Anysphere, a GitHub Copilot rival, has also raised $60 million in Series A funding at a $400 million valuation.

The success of these startups can be attributed to the popularity of GitHub’s AI coding tool Copilot, which has over 1.8 million paying users. The potential productivity gains offered by these tools have been sufficient to ignore their flaws and convince investors to keep pouring money into them. However, it remains to be seen if this trend will continue and for how long.

“Emotion AI” Attracts Investments: The Next Trend for Business Software?

The term “emotion AI” has been gaining traction in recent times, as Venture Capitalists (VCs) and businesses alike are being drawn to its potential applications in the business software space. But is this trend a harbinger of innovation or a recipe for disaster?

Why Home Robots Still Suck

The quest for a reliable and efficient home robot has been ongoing for years, but unfortunately, most attempts have fallen short of expectations. According to Brian, the main reasons behind this failure are pricing, functionality, and efficacy.

Pricing is often a major obstacle, as many home robots come with hefty price tags that are not justified by their performance. The high cost of development and production can make it difficult for companies to sell these robots at a reasonable price, leading to limited adoption and low demand.

Functionality is another area where home robots tend to fall short. Many of them promise the world but fail to deliver on their promises. They may be unable to perform tasks efficiently or effectively, leading to user frustration and disappointment.

Efficacy is also an issue, as many home robots are not designed with the needs of users in mind. They may not be able to adapt to different situations or environments, which can limit their usefulness and make them less effective.

In conclusion, the lack of success in creating a reliable and efficient home robot is due to various factors, including pricing, functionality, and efficacy. To overcome these challenges, companies need to focus on developing robots that are affordable, easy to use, and capable of performing tasks effectively.

Amazon Hires Covariant Founders

Amazon has made a significant move in the robotics space by hiring the founders of robotics AI startup Covariant, along with about a quarter of its employees. Additionally, Amazon has signed a nonexclusive license to utilize Covariant’s AI robotics models.

This strategic acquisition is expected to enhance Amazon’s capabilities in developing and implementing advanced robotic technologies. With Covariant’s expertise and technology, Amazon can further improve its warehouse operations, logistics, and supply chain management.

NightCafe: The OG Image Generator

Before Midjourney, there was NightCafe, one of the original image generators that paved the way for the current AI-generated content revolution. Despite facing moderation challenges, NightCafe is still thriving and kicking.

As a marketplace, NightCafe allows users to create and share AI-generated content, from stunning visuals to thought-provoking art pieces. Its platform has been instrumental in fostering creativity and innovation within the AI community, providing a space for artists and developers to showcase their work and connect with like-minded individuals.

While NightCafe may have faced its fair share of moderation challenges, it continues to be an important player in the world of AI-generated content.

Midjourney Enters the Hardware Realm

Midjourny, a notable competitor to NightCafe, has announced its foray into the world of hardware. This significant development was revealed through a recent post on X, which also mentioned that the company’s new hardware team will be based in San Francisco.

The move marks an exciting expansion for Midjourney, as it continues to push boundaries in the AI-powered art generation space. With this addition, the company is set to further diversify its offerings and explore new avenues of innovation.

Time will tell how Midjourney’s hardware endeavors unfold, but one thing is certain – the landscape of AI-generated art just got a whole lot more interesting!

California’s AI Bill SB 1047 Passes: A Step Towards Preventing AI Disasters?

The California State Legislature has recently passed AI bill SB 1047, aimed at preventing AI disasters. However, Silicon Valley is warning that this move may inadvertently cause one. The bill, which still needs to be signed into law by the governor, aims to improve transparency and accountability in AI development.

What Does the Bill Entail?

SB 1047 seeks to regulate AI development by requiring developers to disclose potential biases and risks associated with their creations. It also aims to ensure that AI systems are transparent and accountable, allowing users to understand how they make decisions.

Silicon Valley’s Concerns

Despite its intentions, Silicon Valley experts are warning that the bill may have unintended consequences. They argue that it could stifle innovation in AI development, as developers may be reluctant to create complex models if they must disclose their inner workings.

What Happens Next?

The fate of SB 1047 now rests with California’s governor. If signed into law, the bill will require significant changes to the way AI is developed and used in the state. However, if vetoed, it may be a setback for those seeking greater transparency and accountability in AI development.

Conclusion

The passing of SB 1047 marks an important step towards regulating AI development in California. While its intentions are noble, Silicon Valley’s concerns highlight the need for careful consideration of the bill’s potential consequences. Only time will tell if this legislation will ultimately prevent AI disasters or inadvertently cause one.

Google Rolls Out Election Safeguards

Google is taking steps to prepare for the upcoming U.S. presidential election by implementing safeguards on more of its generative AI apps and services. As part of these restrictions, most of the company’s AI products will no longer respond to election-related topics.

This move aims to prevent misinformation from spreading through Google’s platforms, which could potentially influence voters’ decisions. The decision is a crucial step in ensuring a free and fair election process, where citizens have access to accurate information.

By restricting responses to election-related queries, Google hopes to minimize the risk of AI-generated content being used to sway public opinion or spread disinformation. This measure will help maintain trust in the democratic process and protect the integrity of the upcoming election.

Apple and Nvidia Could Invest in OpenAI

Nvidia and Apple are reportedly in talks to contribute to OpenAI’s next fundraising round, which could value the ChatGPT maker at $100 billion. This potential investment would be a significant milestone for OpenAI, as it continues to revolutionize the field of artificial intelligence with its cutting-edge technology.

The news comes after OpenAI raised $1 billion in funding earlier this year, valuing the company at around $20 billion. The latest round is expected to bring in even more capital, catapulting OpenAI’s valuation to a staggering $100 billion.

This potential investment from Apple and Nvidia could have significant implications for the future of AI research and development. With the support of two industry giants, OpenAI may be able to accelerate its progress and push the boundaries of what is possible with artificial intelligence.

Research paper of the week

Who needs a game engine when you have AI?

GameNGen: An AI System That Can Simulate Classic Games Like Doom

Researchers at Tel Aviv University and DeepMind, Google’s AI R&D division, have made significant progress in developing an AI system that can simulate classic games like Doom. The system, called GameNGen, is capable of generating game scenes in real-time, mimicking the gameplay experience at a frame rate of up to 20 frames per second.

How It Works

GameNGen was trained on extensive footage of Doom gameplay, allowing it to effectively predict the next “gaming state” when a player controls the character in the simulation. This means that the AI system can generate game scenes based on its understanding of the game’s mechanics and physics.

Real-Time Generation

One of the most impressive aspects of GameNGen is its ability to generate game scenes in real-time. This allows for seamless gameplay, with no lag or delay between actions taken by the player and the corresponding responses from the game environment.

A Doom-like level, generated by AI.

GameNGen: The Latest AI Model to Simulate Games

GameNGen, a new AI model, has joined the ranks of other models capable of simulating games. This is not the first time an AI model has been able to generate video game-like content. OpenAI’s Sora can already simulate games, including Minecraft.

In fact, earlier this year, a group of university researchers unveiled an AI that could simulate Atari games. Other notable models include World Models, GameGAN, and Google’s Genie, which creates interactive 2D worlds from a single image.

Sora can render video games, including Minecraft. Simulating Atari games (World Models, GameGAN, and Google’s Genie are other examples of AI models capable of simulating games.)

GameNGen: A Promising Step Towards Procedurally Generated Games

GameNGen is an impressive game-simulating model that showcases its performance capabilities. While it’s not without limitations, such as graphical glitches and the inability to remember more than three seconds of gameplay, it could be a significant step towards creating entirely new types of games.

The concept of procedurally generated games, which involves generating content on the fly through algorithms rather than manual creation, has been gaining traction in recent years. GameNGen’s potential to take this concept to the next level is vast, with possibilities for creating complex and dynamic game worlds that are unparalleled in traditional game development.

While there is still much work to be done to overcome the limitations of GameNGen, its potential as a game-changer (no pun intended) cannot be overstated. As researchers continue to refine and improve the model, it’s likely that we’ll see the emergence of new and innovative game genres that were previously unimaginable.

Model of the week

The Rise of AI in Weather Forecasting

Devin Coldewey has previously discussed how AI is transforming the field of weather forecasting, from simple short-term predictions to long-range projections spanning centuries.

With its ability to process vast amounts of data and identify patterns, AI is increasingly being used to improve weather forecasts. Gone are the days when users had to rely on manual observations or limited computer models for predicting the weather. Today, AI-powered systems can provide accurate forecasts for varying time scales, from hourly outlooks to 10-day predictions and even century-level projections.

As AI continues to advance in this field, it’s likely that we’ll see further improvements in the accuracy and reliability of weather forecasting, ultimately benefiting individuals, communities, and industries alike.

Aurora: The Latest Breakthrough in Weather Forecasting

Aurora, a cutting-edge model developed by Microsoft’s AI research organization, has made its debut in the world of weather and climate prediction. This innovative technology is trained on a diverse range of datasets related to weather and climate, allowing it to be fine-tuned for specific forecasting tasks with relatively little data.

With its unique capabilities, Aurora promises to revolutionize the field of meteorology, providing accurate and reliable forecasts that can help us better understand and prepare for extreme weather events. Its potential applications are vast, from improving our daily lives to informing critical decision-making processes in industries such as agriculture, transportation, and emergency services.

As a result, Aurora is an exciting development that has the potential to reshape the way we approach weather forecasting and climate prediction. With its ability to adapt quickly and accurately, this technology has the power to make a significant impact on our understanding of the world around us.

Microsoft’s Aurora Machine Learning Model Predicts Atmospheric Variables

Aurora is a machine learning model developed by Microsoft that can accurately predict atmospheric variables such as temperature. According to the model’s GitHub page, it provides three specialized versions: one for medium-resolution weather prediction, another for high-resolution weather prediction, and a third for air pollution prediction.

These versions cater to different needs in the field of meteorology and environmental monitoring, allowing researchers and professionals to utilize the model for various applications. By leveraging machine learning capabilities, Aurora offers a powerful tool for understanding and predicting atmospheric phenomena, ultimately contributing to improved forecasting and decision-making processes.

Aurora’s Performance: A Balance Between Accuracy and Limitations

Aurora’s performance appears quite good relative to other atmosphere-tracking models. It can produce a five-day global air pollution forecast in under a minute, as well as a ten-day high-resolution weather forecast.

However, Aurora is not immune to the hallucinatory tendencies of other AI models. Despite its capabilities, it can still make mistakes. As such, Microsoft advises against using Aurora for planning operations, emphasizing its limitations and potential errors.

AI Data Labeling Startup Lays Off Scores of Annotators

Inc. recently reported on a shocking development in the world of artificial intelligence (AI). Scale AI, a startup focused on AI data labeling, has laid off numerous annotators without warning. These annotators were responsible for labeling training datasets used to develop AI models.

The sudden and unexpected layoffs have sent shockwaves through the industry, leaving many wondering about the implications for the future of AI development. As the demand for high-quality labeled data continues to grow, it remains to be seen how this decision will impact the overall landscape of AI research and innovation.

Layoffs at Scale AI: Rumors and Uncertainty

The tech industry is abuzz with rumors of layoffs at Scale AI, a leading artificial intelligence company. While no official announcement has been made by the company, sources close to the matter have revealed that hundreds of employees may have been let go.

Rumored Layoffs

One former employee who spoke to Inc. shared information about the alleged layoffs, stating that hundreds of employees were affected. However, Scale AI disputes this claim, leaving many wondering what’s true and what’s not.

The Wait Continues

As the situation unfolds, one thing is certain – we’ll have to wait for an official statement from Scale AI before knowing the full extent of the layoffs. Until then, speculation will continue to swirl around the tech community.

The Reality of Annotators Working for Scale AI

Scale AI annotators are not directly employed by the company. Instead, they are hired through one of Scale’s subsidiaries or third-party firms, which can lead to less job security. Labelers often face uncertainty, as they may go extended periods without receiving work or be unexpectedly removed from the platform.

In recent cases, contractors in Thailand, Vietnam, Poland, and Pakistan have been affected by this issue.

Scale’s Recent Layoffs: A Clarification on the Affected Employees

According to a Scale spokesperson speaking with TechCrunch, the company hires contractors through a partner called HireArt. The individuals who lost their jobs were actually employees of HireArt and received severance and COBRA benefits until the end of the month from HireArt.

The layoffs affected less than 65 people last week, as part of a strategy to scale its contracted workforce in line with the company’s evolving operating model over the past nine months. In total, fewer than 500 employees have been laid off in the United States.

Clarification on Recent Layoffs at Scale AI

It may be challenging to decipher the exact meaning behind Scale AI’s carefully crafted statement regarding recent layoffs. Nevertheless, we are actively investigating the matter.

If you were a former employee of Scale AI or a contractor who was recently let go, please feel free to reach out to us in whatever manner is most comfortable for you.