AI chatbots unable to accurately summarise news, BBC finds
-
What temperature and sampling settings? Which models?
I've noticed that the AI giants seem to be encouraging “AI ignorance,” as they just want you to use their stupid subscription app without questioning it, instead of understanding how the tools works under the hood. They also default to bad, cheap models.
I find my local thinking models (like QwQ or Deepseek 32B) are quite good at summarization at a low temperature, which is not what these UIs default to. Same with “affordable” API models (like base Deepseek). But small Gemini/OpenAI API models are crap, especially with default sampling, and Gemini 2.0 in particular seems to have regressed.
My point is that LLMs as locally hosted tools are neat, but how corporations present them as magic cloud oracles is like everything wrong with tech enshittification in one package.
They were actually really vague about the details. The paper itself says they used GPT-4o for ChatGPT, but apparently they didnt even note what versions of the other models were used.
-
Whoops, yeah, should have linked the blog.
I didn't want to link the individual models because I'm not sure hybrid or pure transformers is better?
Looks pretty interesting, thanks for sharing it
-
Turns out, spitting out words when you don't know what anything means or what "means" means is bad, mmmmkay.
It got journalists who were relevant experts in the subject of the article to rate the quality of answers from the AI assistants.
It found 51% of all AI answers to questions about the news were judged to have significant issues of some form.
Additionally, 19% of AI answers which cited BBC content introduced factual errors, such as incorrect factual statements, numbers and dates.
Introduced factual errors
Yeah that's . . . that's bad. As in, not good. As in - it will never be good. With a lot of work and grinding it might be "okay enough" for some tasks some day. That'll be another 200 Billion please.
Is it worse than the current system of editors making shitty click bait titles?
-
Turns out, spitting out words when you don't know what anything means or what "means" means is bad, mmmmkay.
It got journalists who were relevant experts in the subject of the article to rate the quality of answers from the AI assistants.
It found 51% of all AI answers to questions about the news were judged to have significant issues of some form.
Additionally, 19% of AI answers which cited BBC content introduced factual errors, such as incorrect factual statements, numbers and dates.
Introduced factual errors
Yeah that's . . . that's bad. As in, not good. As in - it will never be good. With a lot of work and grinding it might be "okay enough" for some tasks some day. That'll be another 200 Billion please.
Do you dislike ai?
-
Turns out, spitting out words when you don't know what anything means or what "means" means is bad, mmmmkay.
It got journalists who were relevant experts in the subject of the article to rate the quality of answers from the AI assistants.
It found 51% of all AI answers to questions about the news were judged to have significant issues of some form.
Additionally, 19% of AI answers which cited BBC content introduced factual errors, such as incorrect factual statements, numbers and dates.
Introduced factual errors
Yeah that's . . . that's bad. As in, not good. As in - it will never be good. With a lot of work and grinding it might be "okay enough" for some tasks some day. That'll be another 200 Billion please.
alternatively: 49% had no significant issues and 81% had no factual errors, it's not perfect but it's cheap quick and easy.
-
This post did not contain any content.
You don't say.
-
alternatively: 49% had no significant issues and 81% had no factual errors, it's not perfect but it's cheap quick and easy.
It's easy, it's quick, and it's free: pouring river water in your socks.
Fortunately, there are other possible criteria. -
Funny, I find the BBC unable to accurately convey the news
Dunno why you're being downvoted. If you're wanting a somewhat right-wing, pro-establishment, slightly superficial take on the news, mixed in with lots of "celebrity" frippery, then the BBC have got you covered. Their chairmen have historically been a list of old Tories, but that has never stopped the Tory party of accusing their news of being "left leaning" when it's blatantly not.
-
That response doesn't make sense. Please clarify.
A human can move, a car can move. a human can't move with such speed, a car can. The former is qualitative difference how I meant it, the latter quantitative.
Anyway, that's how I used those words.
-
Turns out, spitting out words when you don't know what anything means or what "means" means is bad, mmmmkay.
It got journalists who were relevant experts in the subject of the article to rate the quality of answers from the AI assistants.
It found 51% of all AI answers to questions about the news were judged to have significant issues of some form.
Additionally, 19% of AI answers which cited BBC content introduced factual errors, such as incorrect factual statements, numbers and dates.
Introduced factual errors
Yeah that's . . . that's bad. As in, not good. As in - it will never be good. With a lot of work and grinding it might be "okay enough" for some tasks some day. That'll be another 200 Billion please.
It found 51% of all AI answers to questions about the news were judged to have significant issues of some form.
How good are the human answers? I mean, I expect that an AI's error rate is currently higher than an "expert" in their field.
But I'd guess the AI is quite a bit better than, say, the average Republican.
-
Turns out, spitting out words when you don't know what anything means or what "means" means is bad, mmmmkay.
It got journalists who were relevant experts in the subject of the article to rate the quality of answers from the AI assistants.
It found 51% of all AI answers to questions about the news were judged to have significant issues of some form.
Additionally, 19% of AI answers which cited BBC content introduced factual errors, such as incorrect factual statements, numbers and dates.
Introduced factual errors
Yeah that's . . . that's bad. As in, not good. As in - it will never be good. With a lot of work and grinding it might be "okay enough" for some tasks some day. That'll be another 200 Billion please.
I'll be here begging for a miserable 1 million to invest in some freaking trains qnd bicycle paths. Thanks.
-
This post did not contain any content.
But the BBC is increasingly unable to accurately report the news, so this finding is no real surprise.
-
Do you dislike ai?
I don't necessarily dislike "AI" but I reserve the right to be derisive about inappropriate use, which seems to be pretty much every use.
Using AI to find pertoglyphs in Peru was cool. Reviewing medical scans is pretty great. Everything else is shit.
-
alternatively: 49% had no significant issues and 81% had no factual errors, it's not perfect but it's cheap quick and easy.
Flip a coin every time you read an article whether you get quick and easy significant issues
-
A human can move, a car can move. a human can't move with such speed, a car can. The former is qualitative difference how I meant it, the latter quantitative.
Anyway, that's how I used those words.
Ooooooh. Ok that makes sense. Correct use of words, just was not connecting those dots.
-
They are, however, able to inaccurately summarize it in GLaDOS's voice, which is a point in their favor.
Yeah, out of all the generative AI fields, voice generation at this point is like 95% there in its capability of producing convincing speech even with consumer level tech like ElevenLabs. That last 5% might not even be solvable currently, as it's those moments it gets the feeling, intonation or pronunciation wrong when the only context you give it is a text input.
Especially voice cloning - the DRG Cortana Mission Control mod is one of the examples I like to use.
-
How could I blindly trust anything in this context?
Y'know, a lot of the hate against AI seems to mirror the hate against Wikipedia, search engines, the internet, and even computers in the past.
Do you just blindly believe whatever it tells you?
It's not absolutely perfect, so it's useless.
It's all just garbage information!
This is terrible for jobs, society, and the environment!
-
This post did not contain any content.
I learned that AI chat bots aren't necessarily trustworthy in everything. In fact, if you aren't taking their shit with a grain of salt, you're doing something very wrong.
-
This post did not contain any content.
BBC is probably salty the AI is able to insert the word Israel alongside a negative term in the headline
-
Ooooooh. Ok that makes sense. Correct use of words, just was not connecting those dots.
The roadblock, to my understanding (data science guy not biologist), is the time it takes to discover these things/how long it would take evolution to get there. Admittedly that’s still somewhat quantitative.
Yes.
But it’s the nature of AI to remain within (or close to within) the corpus of knowledge they were trained on.
That's fundamentally solvable.
I'm not against attempts at global artificial intelligence, just against one approach to it. Also no matter how we want to pretend it's something general, we in fact want something thinking like a human.
What all these companies like DeepSeek and OpenAI and others are doing lately, with some "chain-of-thought" model, is in my opinion what they should have been focused on, how do you organize data for a symbolic logic model, how do you generate and check syllogisms, how do you, then, synthesize algorithms based on syllogisms ; there seems to be something like a chicken and egg problem between logic and algebra, one seems necessary for the other in such a system, but they depend on each other (for a machine, humans remember a few things constant for most of our existence). And the predictor into which they've invested so much data is a minor part which doesn't have to be so powerful.
