Kai Riemer’s Post

Good on them. This will lead to higher quality submissions and quicker decisions. Reviewers’ role is to ascertain the quality of my paper, not try to cowrite (aka develop) it. Too many papers in our field are published years after they would have been relevant. MISQ recently published a paper on Enterprise Social Networks the same week Facebook scrapped its Workplace product. Insights that are 10 years late.

AI can't reason. LLMs are language models, and we shouldn't expect them to reason or show human-like cognition. They function completely differently to how humans cognise. What we need to understand is how they work and what they are best at. As a tool that we work with - not as a thing that might replace us. Here is a little test I did. The three screenshots show 1) a typical reasoning test that is frequently used to benchmark LLMs; 2) the output of Claude 3 Opus on exactly that task - which it nails. Perfect. It seems to reason... 3) A quick change of the genders, and replacing of numerals with word-numbers, and - what? Have a read, it is entertaining nonsense. The point is not that it gets things wrong. Yes, yes, humans get things wrong too. The point is that it is a language model that does token prediction. It does not reason. In the meantime, let's work at making LLMs useful as a tool. There is enough there without "searching for a spark". #ai #genai #llms #reason

"AI is disruptive, it will transform the economy." / "AI is over-hyped, there will be disappointment." You know, both those statements can be true. There is no contradiction between the likelihood that AI will have transformative effects across many industries, and the likelihood that we are in for disappointment because of over-blown expectations. Here are two reasons: 1) Transformation happens in workplaces, in markets, in business processes, in business models. Hype and over-blown expectations happen in the share-market, in VC funding, in the start-up sector, and most of all, in the media. The media loves a good hype, stories of super-human intelligence, and it loves a good crash even more. 2) Hype is short-term & crash is short-term. But the transformation happens slowly, steadily and in the long-term. And no, it won't be from AGI or super-human intelligence, but from us slowly figuring out how to make this new computing paradigm useful. So when, inevitably, the hype ends, and the media turns negative on AI, don't forget that your competitors might still be chipping away at that transformation, upskilling their staff, rethinking their processes and services... #ai #generativeai #hype #transformation #upskilling https://lnkd.in/ggVxmX2Y

‘AI still makes mistakes’ – in my view one of the most annoying things to read about AI... [This is the first in a mini-series of posts I intend write about my main bugbears with common views about AI. These are things that annoy me when I read them, and I will explain why.] So, what bugs me about ‘AI still makes mistakes’? Well, it suggests that AI is somehow deficient and that it must be ‘fixed’. And that it’s only a matter of time until it's fixed and flawless (after all, it’s AI, right?). This thinking is quite prevalent in public discourse. Yet, it ignores the very nature of AI. Unlike traditional computing, which is precise and accurate, AI is probabilistic. It encodes patterns from data and then makes predictions, either in the form of classifications (like with image recognition), or in the form of content generation. But predictions can, and always will to some degree be ‘wrong’. In principle! No matter how much fine-tuning, retraining or whatever computational improvements we throw at the technology, it will not change its nature as a probabilistic technology. There will always be some misclassifications, inaccuracies, or ‘hallucinations’. But let’s not forget, it is the probabilistic nature that makes AI so powerful, with abilities that can exceed humans on a range of tasks. We just need to be careful where we employ it, and put safeguards in place when decisions have severe consequences, like keeping humans closely in the loop. And there’s a final thing that bugs me – the notion of ‘still’ implies that it might be better to wait until the technology is perfect, until its somehow no longer ‘beta’. Well, such a waiting game can risk falling behind, when industries and competitors are already adopting AI for productivity and transformation. AI will still make mistakes, regardless of how powerful it might become. And that’s ok, when we understand its nature, how to manage it and what structures to build around it. AI fluency is key. #AI #generativeai #genai

Acceptance rates go up when LLMs are used to "peer" review papers. This is not surprising. Given that LLMs are fine-tuned for positivity, any uncritical outsourcing of reviewing to these systems will result in more positive outcomes. I assume that whenever we can pick up that AI was used, then substantial parts of a review are written by the LLM, which will bring about such results. My own experience of checking my reviews of papers against LLMs is that the outputs are often uncritical and way too positive, as they do lack the kind of nuanced understanding required to pick up on the novelty of ideas (or lack thereof), which again is not that surprising. This can be alleviated somewhat with more sophisticated prompting, but the tendency to be positive remains in my view. Overall these findings should concern all of us. GenAI and LLMs have a place in reviewing - as assistants for certain tasks, but not as reviewers. #LLM #AI #academia #reviewing

Where will AI take us next? Should we be worried? Should we be excited? I'm happy and proud to announce that Sandra Peter and I will be back on stage at SXSW Sydney this October! Thanks to everyone who voted for us to make this possible. Tickets for Southby Sydney are now available with early bird pricing running until 16 May. More on our session and ticketing link here: https://lnkd.in/d_6HQZEH #sxsw #ai #genai #thefutureofai

We tend to forget. The iPhone was a failure when it was released. It didn't even sell much. The iPad came at the height of the Netbook craze (anyone remember?). It was regarded a failure, seen by many as an iPhone that is too big, merely for consuming content, but not much else, because it didn't have a keyboard (or a stylus). The Apple Watch was a failure. Who would want to strap an iPhone to their wrist? What was it good for anyway? Today, it is the most successful wristwatch - ever. And a bona fide health device. The Vision Pro is a failure, because we fail to understand what it is. The cycle continues. Innovation requires experimentation, iteration, not in a lab, but out in the world. We don't know yet what spatial computing might become. Fortunately, the people at Apple are vision pros (sorry :-) - they have the vision and grit to stick it out and evolve design, application, use and marketing to discover over time the potential of new technologies. All this doesn't mean that the Vision Pro as a product or category will be successful. But it sure means that it is way too early to call either way. Patience. #innovation #design #apple

SXSW Sydney 2024 is here. Last year we told you AI was coming for your job. This year, Sandra Peter and I will show you what to know next... and a whole bunch of weird and wonderful AI stuff. Please vote for our session, and meet us in October. You will need to register and create an account (please...). 1) Register for a free account: https://lnkd.in/gedSRMYd 2) Log in and click 'Judge' on the top navigation bar.  3) Use the search bar and open filter to expand options. Enter "AI and You" and cast your vote, or use this link: https://lnkd.in/g3ZvPbZX I know it's cumbersome, but it will be worth it (in October :-). #AI #generativeai #skills