This week, two startups that develop code generation and suggestion tools – Magic and Codeium – raised a combined nearly half a billion dollars in AI funding. That was a giant amount even by AI sector standards, especially considering that Magic has yet to launch a product or generate revenue.
So why the passion amongst investors? Well, programming will not be a simple – or low cost – business. And each firms and individual developers are in search of ways to streamline the more tedious processes surrounding programming.
One Opinion pollthe common developer spends almost 20% of his work week maintaining existing code as a substitute of writing something recent. In a separate studyCompanies reported that they lose $85 billion annually in missed opportunities on account of excessive code maintenance (including fixing technical debt and fixing poorly performing code).
Many developers and corporations imagine that AI tools may also help here. And consultants agree. In a study for 2023 reportMcKinsey analysts wrote that using AI coding tools, developers are able to write down recent code in half the time and optimize existing code in about two-thirds the time.
However, AI programming will not be a panacea. The McKinsey report also found that certain, more complex workloads – equivalent to those who require familiarity with a particular programming framework – don’t necessarily profit from AI. According to the report's co-authors, even junior developers were needed to finish some tasks with AI, versus those without.
“Participant feedback shows that developers actively iterated the tools to realize (high) quality. This suggests that the technology is best used to support developers relatively than replace them,” the co-authors wrote, making it clear that AI will not be an alternative choice to experience. “Ultimately, to keep up code quality, developers need to know the attributes that make up quality code and encourage the tool to provide the suitable results.”
AI coding tools even have unresolved security and IP-related issues. Some evaluation shows that the tools have led to: more faulty code pushed into codebases in recent times. Code generating tools trained on copyrighted code are actually caught render that code in a particular way, which creates a liability risk for the developers who use it.
However, this doesn’t diminish the passion of developers – and their employers – for AI programming.
The majority of developers (over 97%) said in a 2024 GitHub survey that they’ve adopted AI tools in some form. According to the identical survey, 59% to 88% of firms encourage – or now allow – the usage of assistive programming tools.
Therefore, it will not be too surprising that the AI coding tools market could reach a worth of around $27 billion by 2032 (per Polaris Research) – especially if, as Gartner predicts75% of enterprise software developers will use AI coding assistants by 2028.
The market is already hot. Generative AI coding startups knowledgePoolside and Anysphere have accomplished mammoth rounds up to now yr – and GitHub’s AI coding tool Copilot has over 1.8 million paying usersThe productivity gains the tools could deliver were enough to persuade investors – and customers – to disregard their shortcomings. But we are going to see if the trend continues – and for a way long, exactly.
News
“Emotion AI” attracts investments: Julie writes about how some enterprise capitalists and corporations are gravitating toward “emotion AI,” the more sophisticated version of sentiment evaluation, and the issues this might bring.
Why home robots are still crap: Brian investigates why many attempts at home robots have failed so spectacularly. It's all the way down to price, functionality and effectiveness, he says.
Amazon hires founding father of Covariant: On the topic of robots: Last week, Amazon The founders of the robotics startup Covariant together with “a couple of quarter” of the corporate’s employees. The company also signed a non-exclusive license to make use of Covariant’s AI robotics models.
NightCafe, the OG image generator: Your humble servant has profiled NightCafe, certainly one of the unique image generators and a marketplace for AI-generated content. Despite moderation issues, it continues to be lively and thriving.
In the center of the journey it’s all concerning the hardware: NightCafe competitor Midjourney is entering the hardware industry. The company announced this in a post on X; its recent hardware team will probably be based in San Francisco, it said.
SB 1047 passed: The California state legislature just passed the AI bill SB 1047. Max writes about why some are hoping the governor won't sign it.
Google introduces election protection measures: Google is preparing for the U.S. presidential election by introducing safeguards for more of its generative AI apps and services. As a part of the restrictions, a lot of the company's AI products is not going to reply to election-related topics.
Apple and Nvidia could spend money on OpenAI: Nvidia and Apple are According to reports in talks to contribute to OpenAI's next round of funding – a round that would boost the ChatGPT maker's value to $100 billion.
Research paper of the week
Who needs a game engine when you may have AI?
Researchers at Tel Aviv University and DeepMind, Google's artificial intelligence research and development division, last week unveiled GameNGen, an AI system that may simulate the sport Doom at as much as 20 frames per second. The model was trained on extensive footage of Doom gameplay and may effectively predict the following “game state” as a player “controls” the character within the simulation. It is a real-time generated game.
GameNGen will not be the primary model to do that. OpenAI's Sora can simulate games, including Minecraft, and a bunch of university researchers unveiled an Atari game-simulating AI earlier this yr. (Other models of this type range from World models To GameGAN and Google's own genius.)
But GameNGen is probably the most impressive attempts at game simulation to this point by way of its performance. The model will not be without major limitations, namely graphical glitches and the lack to “remember” greater than three seconds of gameplay (which implies GameNGen cannot actually create a functional game). But it could possibly be a step towards entirely recent sorts of games – like procedurally generated games on steroids.
Model of the week
As my colleague Devin Coldewey has written previously, artificial intelligence is taking on the realm of weather forecasting, from a fast query like “How long will this rain last?” to a 10-day forecast to century-level predictions.
Aurora, certainly one of the most recent models available on the market, is the product of Microsoft's AI research organization. Aurora has been trained on various weather and climate datasets and could be tuned to specific forecasting tasks with relatively little data, Microsoft claims.
“Aurora is a machine learning model that may predict atmospheric variables equivalent to temperature,” says Microsoft explained on the model's GitHub page. “We offer three specialized versions: one for medium-resolution weather forecasting, one for high-resolution weather forecasting, and one for air pollution forecasting.”
Aurora's performance appears to be quite good in comparison with other atmosphere-tracking models. (In lower than a minute, it may well produce a five-day global air pollution forecast or a 10-day high-resolution weather forecast.) But it will not be proof against the hallucinatory tendencies of other AI models. Aurora could make mistakes, which is why Microsoft warns that it shouldn’t be “utilized by people or firms to plan their operations.”
Grab bag
Last week, Inc. reported that Scale AI, the AI data labeling startup, has laid off quite a few annotators – people accountable for labeling the training data sets used to develop AI models.
There has been no official announcement on the time of publication, but a former worker told Inc. that as many as tons of were laid off. (Scale AI denies this.)
Most annotators who work for Scale AI should not directly employed by the corporate. Rather, they’re hired by a Scale subsidiary or third-party company, which offers them less job security. Labelers sometimes go without work for long periods of time. Or they’re unceremoniously kicked off Scale's platform, as has happened to contractors in Thailand, Vietnam, Poland and Pakistan. recently.
Regarding last week's layoffs, a Scale spokesperson told TechCrunch that the corporate hires contract employees through an organization called HireArt. “These individuals (those that lost their jobs) were employees of HireArt and were receiving severance and COBRA advantages from HireArt through the tip of the month. Less than 65 people were laid off last week. We have been build up this contract workforce and growing it to size as we’ve got evolved our operating model over the past nine months. Less than 500 were laid off within the United States.”
It's a bit difficult to parse exactly what Scale AI means by this fastidiously worded statement, but we're looking into it. If you're a former Scale AI worker or a contractor who was recently laid off, contact us in whatever way works best for you.
