Whether to raise money, placate shareholders, or generate positive press, AI’s biggest companies have a habit of announcing advancements that are nowhere near ready to ship.
The industry term for this is “vaporware,” defined as products that arrive much later than initially anticipated or in some cases not at all. The AI industry has puffed out plenty of vaporware over the last year, making it tough to distinguish hype from reality.
As we head into 2025, let’s look at the biggest AI products and promises that haven’t quite panned out yet:
Alexa’s generative AI overhaul
Back in September 2023, Amazon announced a “smarter and more conversational Alexa” powered by large language models. A future version of the voice assistant, Amazon claimed, would not only allow for more naturally flowing conversations, but would understand nonverbal cues such as body language and eye contact using the cameras built into certain Echo devices. Once this optional Alexa overhaul arrived, Amazon suggested that you’d have to pay for it.
More than a year later, the “future of Alexa” is in limbo. A report by Bloomberg in October said that beta testers were unhappy with the new version, which gave stiff, long-winded responses and failed to work with existing smart home integrations. Reuters also reported on problems with Amazon’s in-house language models, prompting the company to seek help from Anthropic’s Claude. Having scrapped plans to debut the new Alexa in October, Amazon is reportedly targeting a 2025 launch now.
OpenAI’s GPT-5
In fairness, OpenAI never officially said that it would release GPT-5 in 2024, but CEO Sam Altman confirmed in late 2023 that the company had started working on it. In March, Business Insider cited unnamed sources in claiming that the new large language model would arrive around midyear, quoting an unnamed CEO who called it “materially better” after reportedly seeing a demo.
But by midyear, Altman was already tempering expectations, saying that “we still have a lot of work to do” on GPT-5, and in October he told an audience on Reddit not to expect the new model this year. The company instead released GPT-o1, which it claims can process queries more thoroughly but is also much slower and more expensive.
Claude 3.5 Opus
OpenAI isn’t alone in pushing back new large language models. As recently as October, Anthropic’s model documentation page was promising a 2024 launch for Claude 3.5 Opus, a new version of its largest language model. Now, that same page doesn’t mention Opus at all.
Dario Amodei, Anthropic’s cofounder and CEO, told Lex Fridman last month that “the plan is still to have a Claude 3.5 Opus,” but didn’t provide any timing. And while there’s some talk of Anthropic just using 3.5 Opus to feed synthetic data to its cheaper models, it could also be another sign of diminishing returns for cutting-edge models.
Apple Intelligence’s Personal Context
Look back at the first press release for Apple Intelligence from June, and you’ll notice a big focus on “personal context.” The phrase appears three times in the announcement’s first two paragraphs, hinting at a version of Siri that can dig up information from emails, texts, notes, and more. “Our unique approach combines generative AI with a user’s personal context to deliver truly helpful intelligence,” CEO Tim Cook proclaimed at the time.
Although Apple Intelligence arrived in October, the personal context is still missing. Apple’s opted to stagger its AI feature releases over the next year, and Siri’s contextual powers didn’t make the cut for 2024. It’s still unclear exactly when those capabilities will arrive.
Multi-step reasoning in Google Search
During its I/O conference in May, Google said it would soon bring “multi-step reasoning capabilities” to Search’s AI Overviews, allowing users to ask complex questions with nuances and caveats. For instance, you could ask “Find the best yoga or pilates studios in Boston and show me details on their intro offers, and walking time from Beacon Hill,” and Google’s Gemini AI would pull all of that info together into a series of info boxes.
Google now says those capabilities will have to wait for the Gemini 2.0 version of AI Overviews, which just entered limited testing. Supposedly you’ll start seeing multi-step reasoning in search results early next year.
Google’s Project Astra
Google’s biggest AI-related I/O reveal was “Project Astra,” a prototype app that can identify what you’re looking at in real-time, answer back-and-forth questions about it, and remember key points of what you’re talking about over a 10-minute conversation. In other words, it’s supposed to be a usable version of last year’s Gemini AI demo video that turned out to be staged.
Only you still can’t use it today. Although Google originally planned to launch something like Astra this year, it’s pushed the timeframe into 2025. And while Google is working with Samsung on an augmented reality headset and has given on-rails demos to the press, there’s no price or release date on that product either.
Samsung’s and LG’s AI home robots
Last January’s CES trade show was great venue for demonstrating the desperation of electronics vendors to position themselves as AI leaders. Cases in point are Samsung and LG, which both showed off competing smart home robots with AI capabilities. A ">teaser video for Samsung’s Ballie, for instance, showed the robot alerting its owner via text message that her dog was making a mess, then responding to instructions to give the dog a snack and put on its favorite video. LG promised similar capabilities for its “Zero Labor Home” Smart Home AI Agent, touting its ability to “understand context and intentions as well as actively communicate with users.”
Like so much other CES vaporware, neither of these robots have materialized as actual products you can buy, but we shouldn’t be surprised; Samsung has been talking about some version of Ballie without shipping it for nearly five years now.
Accedi per aggiungere un commento
Altri post in questo gruppo
On a typical day, you can’t turn on the news without hearing someone say that Congress is broken.
When news broke that the United Healthcare CEO was shot in broad daylight early last month, outrage erupted online. But it wasn’t aimed at the assassin. Instead, it was directed at the broken U.S.
Ashley Abramson first came across Sophie Cress in a cold pitch to her work email. Cress was asking to be an expert source for any stories Abramson was working on as a freelance reporter. “I’ve got
Threads, Meta’s X and Bluesky rival, is testing ads with certain brands in the United States and Japan, the company said Friday.
“We know there will be plenty of feedback abo
Sooner or later, the politicians who most admire Donald Trump begin to emulate him. They
