Rob Bowley

@GergelyOrosz @Pragmatic_Eng It would be great to see you posting more on BlueSky. See you have an account. A lot of folks have moved over in the last few weeks (UK, Brazil). You can use buffer.com to make it easier to cross post

So cool: Substack added a sidebar "table of contents" for all articles! Especially useful for @Pragmatic_Eng ones that tend to be longer, and have a clear structure. Works for all existing articles on the platform! Just tap on the left sidebar to open (it's hidden by default!)

I'd you're personally looking for a last straw to leave this platform 👇

Not tried it yet as on hols chromewebstore.google.com/detail/sky-fol…

Enjoying setting up home on BlueSky. Very much has the feels of early Twitter. Lots of activity with other people moving over too (esp UK folks) Have been recommended the Sky Bridge Chrome extension for finding and migrating ppl you follow here (link in reply)

🧵What do the public think about the riots and those taking part in them? A reflection of legitimate concerns Or the actions of far right thugs? I've written for @FT today on new @moreincommon_ polling and focus group research into Britain & the riots. ft.com/content/464500…

First impressions of Bluesky - I like it and feel it's worth the effort of building back up over there. Most ppl I follow here have accounts there (see my Bsky follows if you're looking for folks) Let's face it, this place isn't going to get any better....

I've not really been posting on here for a while, used to love this place Slowly migrating to Bluesky, let me know if you're on there (if you're still here, most seem to have gone already) robbowley.bsky.social Also: linkedin.com/in/robertbowley

X added a setting for "we'll take your data to train grok" without any notice and just defaulted to "yes" for everyone. This is BAD.

@MartinDotNet Only "always" if you're experienced enough to make those kind of judgement calls. Many more teams than you'd like to think are lacking the level of experience needed to do so.

There is always a judgement call made about what "not" to test. You never test everything, you assume that some things won't be an issue, or that you have a covering test case/scenario that mitigates it Let those who have never decided not to test something cast the first stone

People who think agi is imminent aren’t overestimating how quickly ai will improve. They are underestimating how complex the world is. Being an expert on deep learning doesn’t necessarily make you an expert on how the world works.

How do the LLMs compare? Leveraging our "code grading" tech to introduce weekly computationally grounded LLM benchmarking... wolfram.com/llm-benchmarki…

Newsletter: Goldman Sachs has called BS on Generative AI, and I believe that it's time that everybody follows suit - generative AI is unreliable, unsustainable, requires an entire rebuild of America's power grid, and is most decidedly not the future. wheresyoured.at/pop-culture/

Java is dead Kubernetes is dead DevOps is dead And now: Serverless is dead... In this context, 'dead' means: mature, stable, well understood by the industry. It's not the hot topic for marketing, devrel, or conferences anymore.

Given the divergence of platforms, I decided to return to blogging for longer form thoughts. To start, I thought I might as well offer an opinion on the question of the day: Is AI A Silver Bullet? lnkd.in/guzFyhUs You will want a coffee; it's long.

What surprises me about talented people in tech is their capacity to be brilliant in one area and entirely delusional in another, even within their domain. For instance, those who invented modern machine learning, fully aware of its mechanics and inherent weaknesses, might still claim, "Superintelligence is near; we need to prepare." I want to scream, "How can you, knowing exactly how it works, believe it will soon consume us?" Yet, they somehow do. A language model, however good, is still a machine learning model—a simple mathematical formula whose parameters are only as good as the dataset used to estimate them. Any dataset, no matter how large, has high-density regions, like news articles or fiction books, and low-density regions, like cutting-edge scientific and medical articles. Thus, the model excels in high-density areas and becomes unreliable in low-density ones. Anyone capable of training an LLM knows this; it's common machine learning knowledge. Yet, they persist in saying "superintelligence is near." So, I wonder if it's just human nature to excel in one area and be utterly mistaken in another, or if they are cold-blooded liars causing panic to profit from scared people?

@GaryMarcus Thanks for replying :)

@robbowley Those things certainly need be involved; the need for neurosymbolic AI has been central to my arguments for decades. See my Next Decade in AI on Arxiv.

What I think many people don’t get is that the problems generative AI that we are seeing now aren’t new; they aren’t random. They are *persistent*, e.g., neural networks have NEVER been reliable at negation, NEVER been reliable at compositional structure, and NEVER been good at things that are not canonical. And as the post below points out, they are endless. The graphics have gotten vastly slicker but the core problems I keep harping on remain unsolved. In a rational world, people would spend more time considering alternative approaches, instead of wishing for magic or scale.

@GaryMarcus The data improvement opportunities now are quality rather than quantity, which will all be very expensive. Either human created (there are probably $bns being spent on this already) or licensed

Why assume that the breakthrough we need now is even MORE data, rather than better algorithms? If we can’t get AI to be solid with an entire internet’s worth of data, are we really doing AI right?

In the late 1960s top airplane speeds were increasing dramatically. People assumed the trend would continue. Pan Am was pre-booking flights to the moon. But it turned out the trend was about to fall off a cliff. I think it's the same thing with AI scaling — it's going to run out; the question is when. I think more likely than not, it already has.

@channingwalton It's the hardest thing to do and takes lots of experience. Also, "smart" engineers (much brighter than me) love to make things complicated, I'd guess because it validates their intelligence. See any time anyone uses reflective or asynchronous programming