Axel Halász’s Post

As Marques Brownlee points out, today's AI products aren't really products at all — companies are asking people to pay large amounts of money for something that does not work as advertised, and which might possibly work at some unknown point in the future, if we're lucky. What a waste. It's so weird to work on the science of tech safety during such a brazen and dramatic race to the bottom in quality. If AI were cars, it would be the equivalent of working on seatbelts when manufacturers were forgetting to add wheels. The sad thing is that AI doesn't have to be this way. When I was at SwiftKey, we shipped working software to hundreds of millions of people in ways that made reliable promises, met practical needs and also improved our machine learning models from the feedback we received. And we were far from unusual. The truth is that today's AI companies and their leaders are *choosing* to ship broken systems, that fail to work as advertised, driven less by engineering than by false hopes in the power of hype, market share, and extractive data practices.

Will standalone AI devices actually take off as a consumer product?

I saw this on YouTube this morning and watched Matthew Berman's review yesterday. Matthew, overall, really likes this, though he notes it is not performing all four of the tasks as advertised. The takeaways I took from both reviews (and I'm still looking forward to receiving mine in June, here in the UK) are that: (Pros) it's smaller than expected, very well made, and answers research questions very quickly (though is not immune to halucinations, it seems). (Cons) it cannot perform most of the functions advertised in January by their founder Jesse, the battery life is ridiculously tiny, there is no back button, and the (TFT) touchscreen can (at this moment in time) only be used for the onscreen keyboard, which is utterly crazy, in my view. What really irks me is that it cannot (yet) interface with your calendar, create tasks/to-dos or send emails. That's very strange for an AI personal assistant. But weigh this against the $200 price tag and the fact that there are no service charges or subscription fees. This is a very affordable gadget/assistant and, as was mentioned in the video you cite above, this is not unlike the method used by Tesla to get huge amounts of real-world training data from early adopters.

These AI device companies have exciting concepts but of the main two the Humane PIn or the Rabbit R1 neither are ready for the market as this review shows. Do people think they should release before the product is ready or wait? #ai #devices https://lnkd.in/e-TJYvDX

Another week, another AI gadget review from Marques Brownlee. This time it's the Rabbit R1. About half way through the video, I thought there was a disconnect between the decent feedback and the title "Barely Reviewable". Compared to the Humane AI Pin, the R1 seems to be a finished and functioning product. Then he dives into the killer features which are partially finished or yet to be released. Around the 14 minute mark, he notes the litany of unfinished products released to market recently. And that was my a ha moment 💡 We saw this movie 11 years ago. A product that was released to market in beta form. The message was developers could pay for the privilege of testing ($1500!) and defining the future of the product. The reality was a solution trying to find a problem and the company completely screwing up the go to market strategy. Remember Google Glass? I'm guessing Google hopes you forgot. What's the connection? The tech industry has swung too far in the direction of "ship it". I absolutely do not subscribe to waiting for perfection. Analysis paralysis is dangerous. But we are seeing an alarming trend of half-baked products being sold at full price as final goods. I love that software-driven gadgets have the ability for upgrades and new features. But when are these killer features going to be released, if ever? Should customers be charged up front and wait? Hardware has a limited life. If a customer is only able to enjoy those killer features for a year or so, do the economics make sense to purchase? Don't expect to be able to sell to the masses with this strategy. Products require a basis on solving customer pain points and needs. If "ship it" is being used as an excuse to deliver junk to the market, those companies should prepare for the backlash. Let's get back to fundamentals instead of a rush to make money. #tech #product #productmanagement #ai https://lnkd.in/gw5jhbwj

Is AI lowering the bar for product quality? How many times over the last year or two have you encountered this situation: on a landing page about a product, spaceships roam the expanses of the universe with AI under the hood, and when you try the product itself, it turns out that there is a foot-powered scooter with the simplest use cases and promises of improvements in future? I realized that I had been in this situation many times. The idea was prompted by a review of the Rabbit R1 gadget. This is a box with AI inside, the manufacturers of which promised a bunch of features. In fact, it can answer simple questions and works clumsily with several applications. At the same time, the developers are asking for the full price, promising advanced features such as voice instructions for working with any application someday later. Well, at least the price is “only” $200, not $700 like the analogue from Humane AI. By the way, Marques Brownlee also has a review of Humane AI. Titled "The worst product I've reviewed... so far" 😄 Apparently, the point is in the early stage of development of AI technologies and their specifics. To make AI work well, you need a lot of data. To get a lot of data, you need many users with their unique behavior scenarios. To get many users, you need a quality product... This turns out to be another chicken and egg problem that many models have. The advice Marques gives at the end of the issue, and I second it, is to choose products based on what they already do, not what their creators promise in the future. An old tip that becomes even more relevant with the widespread use of AI technologies! ▶️ https://lnkd.in/ecvWFtxh #aifuture #gpt #ai #chatgpt #chatgpt4 #aiadoption #artificialintelligence #artificialgeneralintelligence #artificialintelligenceforbusiness #artificialintelligenceai #artificialintelligencefordesign #tech #it #techtips #technologies #skills #techcommunity #tools #freetools 

Here it is another AI product that gets mixed/bad reviews. Certainly not as bad as the UI Pin, the Rabbit R1 is still too early, not yet an MUP (Minimum Usable Product). I am excited for these kind of products that are innovating an oversaturated app-based market, but Marques Brownlee is actually giving a good advice: "Buy a product for what it is now, not for what it promises it will be in the future". That was the whole point with minimum viable (or usable) products. Make a product that solves a well defined huge problem for a specific niche of people, and build on top of that. I think it's ok to sell non-refined products, as long as they are still solving that well defined huge problem. To understand what the MVP is, is one of the first biggest obstacles for every startup. This being said, I would still like to see more companies working on AI products. What about you? #ai #rabbitr1 #aipin

■ Our #Data Is What They're Looking For - Reflecting on #RabbitInc's Strategy and the Implications of Early #AI Releases ■ After watching Marques Brownlee's insightful review of the #RabbitR1, it's clear that RabbitInc's launch strategy is about more than just marketing. By entering the market before slower, larger tech giants, they aim to embed their AI into users' daily lives, collecting vast amounts of personal data to enhance future functionalities. This approach highlights a systemic shift in tech priorities - from selling devices to securing a spot in our daily routines through AI-driven personalization. This strategy, however, raises significant ethical questions about privacy. The era we're entering will harvest data not only on our choices but also our interactions and emotions, pushing the boundaries of personal privacy. As AI technologies like #ChatGPT evolve to retain and utilize our information for deeper personalization, the implications stretch beyond convenience. The necessity of our data in shaping these interactions underscores a pivot towards AI systems that understand us on an unprecedentedly personal level. Yet, this introduces a crucial trade-off: the advantages of personalized AI assistants come at the cost of our privacy. Those willing to share their data might find themselves at a technological advantage, leveraging AI for superior efficiency in daily tasks. Despite these benefits, the potential for a middle ground exists. Keeping personal data localized on individual devices could mitigate broader privacy risks while still harnessing AI's benefits. This solution requires robust dialogue about innovation, user autonomy, and privacy protection. As we navigate this landscape, the choices made by companies like RabbitInc will likely set new norms for privacy and personalization in technology. Balancing these aspects will not only dictate the trajectory of consumer technology but also shape broader societal norms around privacy and data use. https://lnkd.in/dcKQ3uWg

I've done a brief video on SBERT embedding model used with Kernel Memory. It is part of my series about Kernel Memory and Semantic Kernel, you can watch in my youtube channel. #ai #sbert #embeddings https://lnkd.in/dw33jaW5

“It’s become abundantly clear over the course of 2024 that writing good automated evals for LLM-powered systems is the skill that’s most needed to build useful applications on top of these models. If you have a strong eval suite you can adopt new models faster, iterate better and build more reliable and useful product features than your competition” https://lnkd.in/dBmtrvUZ