philipsw.bsky.social
@philipsw.bsky.social
Disappointing. Doesn’t cover the case where clear instructions from AI don’t work and result in an error message - it happens now and then. Also doesn’t cover the case of AI producing apparently working code but with security holes.
February 2, 2026 at 8:08 PM
Disappointing. Doesn’t cover the case where clear instructions from AI don’t work and result in an error message - it happens now and then. Also doesn’t cover the case of AI producing apparently working code but with security holes.
Disappointing article. It doesn’t cover the case where the instructions form AI are clear, but wrong, and the running the code results in error message - happens every little while. Also doesn’t cover dangers of apparently working code with security holes or deliberate malicious side effects.
February 2, 2026 at 8:06 PM
Disappointing article. It doesn’t cover the case where the instructions form AI are clear, but wrong, and the running the code results in error message - happens every little while. Also doesn’t cover dangers of apparently working code with security holes or deliberate malicious side effects.
I find that quite often Newsnight is worth watching, perhaps once every two programmes. Typically (but not always), there are insightful guests with useful analysis and insufficient time to express it.
February 1, 2026 at 2:33 PM
I find that quite often Newsnight is worth watching, perhaps once every two programmes. Typically (but not always), there are insightful guests with useful analysis and insufficient time to express it.
Excellent article indeed.
January 2, 2026 at 12:44 PM
Excellent article indeed.
Can you add further practical examples, for example on how you decide which scientific papers you take seriously? I presume that you can check a paper of mathematical deductions a your area. For a data based paper, what are your heuristics for judging whether the authors are reliable?
December 8, 2025 at 6:20 PM
Can you add further practical examples, for example on how you decide which scientific papers you take seriously? I presume that you can check a paper of mathematical deductions a your area. For a data based paper, what are your heuristics for judging whether the authors are reliable?
It’s great for translating into a language which you know well enough to check the output, but where using AI is a great time saver.
December 3, 2025 at 6:15 PM
It’s great for translating into a language which you know well enough to check the output, but where using AI is a great time saver.
Brits always treasure an eccentric; you could build on that aspect. Anyway, please keep writing.
October 5, 2025 at 7:46 PM
Brits always treasure an eccentric; you could build on that aspect. Anyway, please keep writing.
It’s a tool to be used in full knowledge of its limitations. With that approach, it can do some useful things better than any other tool. The output always needs to be checked.
September 29, 2025 at 7:21 PM
It’s a tool to be used in full knowledge of its limitations. With that approach, it can do some useful things better than any other tool. The output always needs to be checked.
Favourite is a high bar, but much of Red Dwarf was very good.
September 20, 2025 at 2:03 PM
Favourite is a high bar, but much of Red Dwarf was very good.
I’ve just seen a partially eclipsed moon, from the edge of Oxford. A couple of minutes later completely covered by cloud.
September 7, 2025 at 7:48 PM
I’ve just seen a partially eclipsed moon, from the edge of Oxford. A couple of minutes later completely covered by cloud.
With respect, I completely disagree. Physics is all about finding the simplest explanation (mathematical theory) that fits the observed facts. There’s no point in having extra parameters in your theory, that aren’t well determined by observation.
September 5, 2025 at 9:59 AM
With respect, I completely disagree. Physics is all about finding the simplest explanation (mathematical theory) that fits the observed facts. There’s no point in having extra parameters in your theory, that aren’t well determined by observation.
Not many people are described as hoist, except by their own petard.
Few things other than points are moot.
Few things other than points are moot.
August 12, 2025 at 4:12 PM
Not many people are described as hoist, except by their own petard.
Few things other than points are moot.
Few things other than points are moot.
If you want your own T-shirt, design it and get vista print to print one for you - it worked when my wife wanted one.
July 30, 2025 at 4:14 PM
If you want your own T-shirt, design it and get vista print to print one for you - it worked when my wife wanted one.
What a sad loss
July 18, 2025 at 11:01 AM
What a sad loss
I respectfully disagree with you. Keir Starmer has a job to do, however distasteful; it’s not in the UK’s interests gratuitously to add conflict to our relations with the US. If you’re not in government you can take a stand; he has to find a way to work with Trump.
July 4, 2025 at 8:32 AM
I respectfully disagree with you. Keir Starmer has a job to do, however distasteful; it’s not in the UK’s interests gratuitously to add conflict to our relations with the US. If you’re not in government you can take a stand; he has to find a way to work with Trump.
Taking a trained model and tuning it using further training data definitely is feasible.
June 30, 2025 at 11:00 AM
Taking a trained model and tuning it using further training data definitely is feasible.
As someone who is only a service user, it would be great to have some examples of what’s now worse from your perspective.
June 17, 2025 at 10:53 AM
As someone who is only a service user, it would be great to have some examples of what’s now worse from your perspective.
Thanks for the article; you continue to be brilliant - no doubt history on repeat.
June 16, 2025 at 7:58 PM
Thanks for the article; you continue to be brilliant - no doubt history on repeat.
Rules out countries without a cheese culture - much of East Asia.
April 27, 2025 at 9:20 AM
Rules out countries without a cheese culture - much of East Asia.
That’s not surprising. The training data will have included several examples of explaining phrases and almost none of saying “that’s not a well known phrase”. LLMs work by providing probable next word (as derived from training data), not on any concept of truth.
April 26, 2025 at 1:35 PM
That’s not surprising. The training data will have included several examples of explaining phrases and almost none of saying “that’s not a well known phrase”. LLMs work by providing probable next word (as derived from training data), not on any concept of truth.
As I undergraduate, I was told (and believed) that the rowing club at St John’s college, Cambridge had been re-named following an incident in which a cox was killed during a “bumps” race. My informant was then a cox in the boat club. Years later I learned the story to be a commonly believed myth.
April 17, 2025 at 6:27 PM
As I undergraduate, I was told (and believed) that the rowing club at St John’s college, Cambridge had been re-named following an incident in which a cox was killed during a “bumps” race. My informant was then a cox in the boat club. Years later I learned the story to be a commonly believed myth.
There’s a 3rd category that I find acceptable. Our neighbourhood group messaging exclusively uses a WhatsApp community announcement group. Most messages fall into types a) and b), but I can cope with the few informational messages that don’t.
March 31, 2025 at 5:18 PM
There’s a 3rd category that I find acceptable. Our neighbourhood group messaging exclusively uses a WhatsApp community announcement group. Most messages fall into types a) and b), but I can cope with the few informational messages that don’t.
I really prefer reading. I can get opinions/info from videos, but it’s a less good experience. Given a choice I’ll go for the text.
Biggest gripe with video is that with text I can skim read to past less interesting detail if I wish, not so with video.
Biggest gripe with video is that with text I can skim read to past less interesting detail if I wish, not so with video.
February 20, 2025 at 5:59 PM
I really prefer reading. I can get opinions/info from videos, but it’s a less good experience. Given a choice I’ll go for the text.
Biggest gripe with video is that with text I can skim read to past less interesting detail if I wish, not so with video.
Biggest gripe with video is that with text I can skim read to past less interesting detail if I wish, not so with video.
Really useful, thank you.
February 12, 2025 at 9:36 AM
Really useful, thank you.