jamesfern.bsky.social
Senior Lecturer; SFHEA; Director of Teaching; #Sport and #Exercise #Science #rstats #openscience #bayes #MUFC
https://www.mdpi.com/2227-7102/14/10/1102
221 posts
940 followers
2,278 following
Regular Contributor
Active Commenter
comment in response to
post
Thanks for taking time to produce this critique, I agree with a lot of it. I'm confused by Kosmyna's comment to Time about trying to trick people using LLMs to summarise, as they state the model in the paper clearly...? Am I missing something?
comment in response to
post
Congratulations to you and your team on the huge amount of work that's gone in to this paper. Could you clarify the confusion that seems to be doing the rounds in relation to this comment? In your paper you do state o4 was used by participant so it makes the Time comment confusing. Thanks!
comment in response to
post
it's stated they used o4 in the paper?
comment in response to
post
It's clearly stated in the paper they used o4. What am I missing here?
comment in response to
post
What am I missing here? The paper explicitly states participants used 4o?
comment in response to
post
The fact human cognitive biases and errors persist despite centuries of study doesn't mean they're categorically different from 'solvable' problems. Both systems might be operating correctly according to their architecture, while still producing errors.
comment in response to
post
I'm not sure thats strictly true, ROME technique shows we can locate where specific facts are stored & edit them to change what the model "knows", suggesting internal representations are more interpretable than you imply. Hallucinations persisting reflect complexity rather than unfixability
comment in response to
post
I think it could be argued that (with sufficient technical expertise & access to the model weights) you could 'trace' the cause of an AI error/hallucination (& 'correct' it). Equally, is it fair to say my own correct inferences/outputs are byproducts of materially the same process as my errors?
comment in response to
post
I don’t necessarily disagree, but would be interested in your logic for the claim ‘human error is a very different animal’.
comment in response to
post
I’ve just finished reading John C Lennox ‘2084 and the AI Revolution’.
This position is unashamedly presented as a ‘scientific explanation’ for all the miracles in the bible.
comment in response to
post
The data is pretty reliable. c.50-70% students admit to cheating, we ‘catch’ around 2%. This was prior to GenAI. Latest surveys show c.95% students are using GenAI in some capacity, the line between ‘use’ and cheating is incredibly blurry.
comment in response to
post
Maybe he’ll reach 3000 lost possessions with us 🤦♂️ averaging 550 per season so far, for the lofty return of…8 assists 🤯
comment in response to
post
Oh yes! How could I forget!
Love doing that 24hrs before the actual board so there is no time to actually find any solutions to the problems 👌
comment in response to
post
And have you also sat through hours of exam boards to approve that marking….
And then repeated that to approve each student outcome…
Then repeated that to consider mitigating circumstances…
comment in response to
post
what exactly would Utd be 'replacing'? An average of 8 assists per season? Or someone to give the ball away more (2200 times!) than any other player across ALL of Europe?
comment in response to
post
Consider 'positive' KPIs like assists relative to negative actions, like losing possession. Bruno is outrageously expensive in terms of misplaced passes per assist.
comment in response to
post
Presumably you play Bruno in CM then? And Amad, RWB? Mainoo? Mount? Zirkzee? It fees like we have too many players for those two 10 roles. Would rather see money spent on a striker.
comment in response to
post
Are you really sold on Mbeumo? Apart from this season, he's not scored more than 8 (non-pk) goals in a season since 19/20?
comment in response to
post
Such a shame we haven't managed to remove Bruno, more of the same next year I fear if he stays.
comment in response to
post
Ditto! This looks great!
comment in response to
post
When the player in the middle of the pitch, your apparent leader, gives the ball away nearly 600 times in 35 games, you need to start there.
comment in response to
post
Doesn’t Salah have 20 more goals & twice as many assists?? And Salah also doesn’t give away possession on average 20 times a game to achieve those numbers either. Honestly its hard to overstate how statistically awful he is. And the eye test confirms it, he gave the ball away for their goal tonight.
comment in response to
post
His numbers are really not that impressive relative to his position and number of attempted passes and shots. In 2023/24 Rodri had more assists and non-pen goals than him,while averaging 92% pass completion, compared to Bruno’s 72%. Bruno misplaced a pass 589 times in 35 games.
comment in response to
post
It was literally his two misplaced passes that created their goal.
Bruno is objectively the worst passer of the ball in his position across all top 5 leagues, by a mile! Even when compared relative to assists/goals.
comment in response to
post
Thinking “his numbers” are some kind of excuse is exactly what has got us into this state. By your logic you’d have never started Scholes… 🤦♂️
Bruno is objectively the worst passer of the ball in his position across all top 5 leagues, by a mile! Even when compared relative to assists/goals.
comment in response to
post
The fact no one discusses how utterly horrendous Bruno is, consistently, is like when Rutger Bregman said at Davos, “it’s like being at a firefighters conference and no one is allowed to talk about water”.
He’s the one consistent component of all our worst moments.
comment in response to
post
Am with you. I worry INEOS won’t/cant (financially) accept that.
comment in response to
post
I agree, but that’s only 2 transfer windows. Realistically how many can he sign and what % of those will be successful? He probably needs 6-7 new first team players? Are we and the board okay with another season trying to avoid relegation?