Trouble is, they were friends of my wife, from her aquafit group. She wasn't previously aware they were Jovo's. Usually if I answer the door to other Jovo's, I just politely inform them I am atheist, which is generally sufficient.
Last pair of Jovos (or Mormons, I never gave them the cance to tell me which) that came to mine did so while I was at work. Not being in the best mood at the time, I just snapped "Religion is for the weak of mind!" via the video doorbell before inviting them to depart. It's fun to surprise people who don't know video doorbells are motion-activated, especially the kids who try to sneak up and play "Knock down Ginger" on it.
Here's the thing with the door-to-door missionaries... when you're rude to them, you're helping their organisation. You'll be amazed to hear that door-to-door proselytising doesn't work. Shocker, right? Nobody wants to be interrupted and preached at. Nobody converts. That's not a thing. But that's not actually the point. The point is that you get your starry-eyed eager beavers and send them out door-to-door knowing full well that it doesn't work, and actively hoping people'll get pee'd off and tell 'em to shove the Watchtower where God can't see. They want that to happen. Why? Because it reinforces that the outside world is full of sinful wicked heathens. Horrible people looking to dim God's light. Who will abuse you at the drop of a hat, when all you're trying to do is save them. Why would you go outside? Stay here, in the warm embrace of brotherhood, with the saved, people who understand you, who are kind to you. Forget the rest of the world, you're one of us and always welcome here. It's a brainwashing tool. Every time you tell 'em to feck off and slam the door in their face, you're helping reinforce that message. This ain't a conspiracy theory, this is literally why they do what they do. Anyway, it's more fun to tell 'em that you've been disfellowshipped and watch 'em run away...
A couple of discussions at work lately have made me think a lot more about the general sense of discomfort and unease that’s been building around this job. It isn’t a question of money. Sure - more money would be nice, but when is being paid more ever unwelcome? I don’t feel valued here. There’s an awful lot more I could talk about or point to, but that’s the core of it: I don’t feel like my skills, my contributions, or what I have to offer, are valued. The fact that I’m trying to buy a house at the moment is very much working in my employer’s favour. But… sooner or later, I am going to get past that.
Burned through hours yesterday trying to get an upcoming product working ahead of embargo lift on Thursday, and more today. Turns out there was a software update on the 8th of January which broke it - you have to install an older version from December. Amaze. EDIT: Good god. Yes, yes, I knew it'd get the answer wrong... but that wrong?
That’s… that’s not even “technically correct”… 10/10 for effort there, but 0/10 for actually answering the question… Early last year I played around with local model inferencing, and I’m sure qwen2 was one of the models I tried. It was becoming clear that we were going to start using “AI-assisted development” no matter what, so I was still trying to gently steer us towards local inferencing… An effort that proved entirely in vain, because it didn’t take long for us to put all our eggs in Cursor’s basket and have that be mandated as our primary IDE… but I digress… If I’ve still got any models cached by Ollama - and Open WebUI still works - I’ll give this a try out of sheer curiosity…
You may get better results: these particular models are quantised down to INT4, which I'm sure ain't helping matters.
Strrrrrrrrrrrrrrrrrrrrrrrrawberrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrry is the correct spelling then? Huh, this AI is clever
Other answer-shaped-objects it gave to the same prompt include: The word "strawberry" is a noun and not a number. The currency used in India is Indian Rupee (Rs). However, there is no such thing as a "strawberry." The word "strawberry" contains two Rs and one s, which equals 14 Rs. The word "strawberry" contains a total of 7 letters. Each letter represents one Rs in the English language, with no specific value or unit attached to each letter. Therefore, there are no Rs in the word "strawberry." The word "strawberry" contains two Rs and three other letters: "a, b, t". So the total number of Rs is: 2 (Rs in "straw") + 3 (Rs in "straw") = 5 Rs Therefore, there are five Rs in the word "strawberry". The first time I prompted it, it responded with a bunch of "reasoning" to reach the answer-shaped object "Rs. $16". I hit "explain," and it changed its "reasoning" (it had said the letters had a value 0 to 9, then said they had a value 1 to 8 - which, somehow, didn't change the answer-shaped object at all), then concluded with - and this is a direct quote: "This means that morgans and I will have $16 in our accounts." <sigh>
I asked the same question of various models offered by Duck.ai (free). Most answered a simple 2 or 3, one, Claude Haiku 3.5, answered: Let me help you count the number of Rs in the word "strawberry": There are 2 Rs in the word "strawberry". The Rs are located in the middle of the word: "st**r**awbe**r**y"
Enough electricity to boil a kettle versus a few microwatts to do a simple countif statement and most are still getting it wrong!
Fun fact: the ones that get it right do so only because A) they have specific exceptions in their system prompt for handling this and other common gotchas, and 2) they're trained on data which includes people taking the mickey out of LLMs *not* being able to do it. I've finished my review, and figured - hey, we're always told LLMs are great at summarisation (they're not), so let's get deepseek_r1:1.5b to summarise the review. That'll be fun! [HailoRT] [warning] Conversation context is full. It is adivsable [sic] to clear context as cache size was reached Haa. Chocolate teapot.
The latest Windows update scared/annoyed the poo out of me by completely failing to connect to my TV after installation. Even a couple of restarts and a GPU driver update didn't help. Left everything off for a few minutes and thankfully things are back to normal. Not what I needed just before officially starting work at home for the day.
Got curious and tried a couple this morning, including your very first example - I tried to stick to similar parameter sizes, but I have no idea what the quantisation was. qwen2:1.5b: Uhhh… sure… sure, buddy… qwen2.5:1.5b: qwen3:1.7b: Ok, so, this one went wild. It “thought” for a solid 21 seconds before giving an answer - I’ll put it in spoiler tags to avoid clogging things up… Spoiler: Wot qwen3 ‘thought’ “Okay, so I need to figure out how many times the letter 'R' appears in the word "strawberry". Let me start by writing down the word and looking at each letter one by one. First, let's spell out "strawberry". Breaking it down letter by letter: S, T, R, A, W, B, E, R, Y. Wait, let me check that again. The word is "strawberry". Let me confirm the letters. S-T-R-A-W-B-E-R-R-Y? Or is it S-T-R-A-W-B-E-R-Y? Hmm, maybe I need to count carefully. Let me write it out again: S, T, R, A, W, B, E, R, R, Y. Oh, wait, so the letters are S, T, R, A, W, B, E, R, R, Y. So that's 10 letters in total. Now, let's check each letter for 'R's. Starting from the first letter: S – not an R. Then T – no. Next is R – that's the first R. Then A, W, B, E, R, R, Y. So after the first R, there's another R at the 7th position? Wait, let me count again. Wait, the letters are: S T R A W B E R R Y So positions 3, 8, and 9 are R's. Wait, position 8 is the first R after the initial R at position 3. Then position 9 is another R. So that's three R's? Wait, let me check again. The word is "strawberry". Let me spell it out properly. S-T-R-A-W-B-E-R-R-Y. So yes, after the first R (position 3), there's another R at position 8, and another at position 9. So that's three R's. But wait, sometimes people might miscount. Let me make sure. Let's go through each letter: S (no), T (no), R (yes, count 1), A (no), W (no), B (no), E (no), R (count 2), R (count 3), Y (no). So three R's. Hmm, but I'm a bit confused because sometimes when you spell words, there might be variations. Let me check if "strawberry" is spelled with two R's or three. Wait, "strawberry" is spelled S-T-R-A-W-B-E-R-R-Y. So yes, two R's in the middle. Wait, no, wait: S-T-R-A-W-B-E-R-R-Y. So between the 'w' and 'y', there are two R's. So that's two R's? Wait, no. Let me count again. Breaking it down: S (1), T (2), R (3), A (4), W (5), B (6), E (7), R (8), R (9), Y (10). So positions 3, 8, and 9 are R's. So three R's. Therefore, the answer should be 3. But wait, I think sometimes people might think that "strawberry" has only two R's. Let me check another way. Let me look up the spelling of "strawberry" to confirm. But since I can't actually look it up, I have to rely on my memory. I remember that "strawberry" is spelled with two R's. Wait, but in the word "strawberry", after the 'w', there's 'b', then 'e', then two R's. So yes, two R's. Wait, but according to the breakdown above, there are three R's. Wait, this is conflicting. Wait, let's spell it again: S-T-R-A-W-B-E-R-R-Y. So between the 'w' and 'y', there's 'b', 'e', then two R's. So that's two R's. But in the breakdown earlier, I had three R's. Wait, no. Let me count again: Positions: S T R A W B E R R Y So positions 3, 8, and 9 are R's. So that's three R's. But if the word is spelled as S-T-R-A-W-B-E-R-R-Y, then between the 'w' and 'y', there are two R's. So that's two R's. Wait, but the positions are 3, 8, 9. So that's three R's. But how? Wait, maybe I'm miscounting. Let me write the letters with numbers: 1: S 2: T 3: R 4: A 5: W 6: B 7: E 8: R 9: R 10: Y So positions 3, 8, 9. So three R's. Therefore, the answer is three. But I thought maybe it's two. Maybe I was confused because sometimes people might miss one. But according to the spelling, it's three R's. So the answer is three. But let me check once more. Let's write the word out: S-T-R-A-W-B-E-R-R-Y. So after the first R (position 3), there's a 'b', then 'e', then two R's. So that's two R's. Wait, but in the spelling, after the 'e', there's an 'r', then another 'r', so that's two R's. So total R's are three? Or two? Wait, if the word is S-T-R-A-W-B-E-R-R-Y, then between the 'w' and 'y', there's an 'e', then two R's. So that's two R's. But the third R is after the 'e'? Wait, no. Let me parse the word again. The word is "strawberry". Let's break it down: Strawberry is S-T-R-A-W-B-E-R-R-Y. So the letters are: S, T, R, A, W, B, E, R, R, Y. So that's 10 letters. So the R's are at positions 3, 8, and 9. Therefore, three R's. Therefore, the answer is three. But I need to make sure that I'm not making a mistake here. Let me think of another way. If I count the letters in the word: S-T-R-A-W-B-E-R-R-Y. So the letters are: S, T, R, A, W, B, E, R, R, Y. So the R's are in the third position, then the eighth and ninth positions. So three R's. Therefore, the answer is three. But I'm a bit confused because sometimes when I spell words, I might miscount. But according to this breakdown, it's three. So I think the correct answer is three." gemma3:4b: This was the smallest version of this model… ministral-3:3b: “The word "strawberry" contains 3 letters”, huh? That’s news to me… deepseek-r1:1.5b: This one also had a little “think” before answering, but this time it only took 2 seconds. Spoiler: Wot deepseek-r1 ‘thought’ “I'll start by examining the word "strawberry" to identify and count all of its uppercase letters. First, I'll look for any capital letters. The first letter is 'S', which is uppercase. Then, checking each subsequent letter: 't', 'r', 'a', 'w', 'n', 'e', 't', and 'r' are also uppercase. I'll count the number of these uppercase letters to determine how many times "Rs" appear in the word. Finally, I'll summarize the total count to provide a clear answer." Wha… what…? And yes, “\boxed{2}” did appear in its answer, I didn’t add that and it’s not a formatting glitch… Well, that was fun, I think I’m going to go and have a little treat now; we should still have a “Strawnetr” or two in the fridge.
Methinks qwen3:1.7b might have been taking the piss, at least I hope so, or is it really that useless?
Nope, that's what the "reasoning" models do. Y'see, all large language models - all large language models - work by turning your input into tokens then selecting the most statistically-likely tokens that continue said input (while adding some randomness into the mix, which is why you can get three different answers when you repeat your question three times). They don't "know," they don't "think," they don't "reason," they just continue the token stream. If your input is in the form of a question, the most likely continuation is in the form of an answer - 'cos the training data was full of questions followed by answers. That's why they give you an answer-shaped output, but it's important to know it's not actually an answer. Even when it's right, it's not actually an answer. It's just a stream of tokens taking the form of an answer-shape, because that's what followed question-shapes in the training data. The "thinking" and "reasoning" models aren't trained any differently, but they are pre-prompted differently. Basically, there's a "system prompt" which says things like "you are a very helpful large language model, eager to assist your users, and definitely won't ever tell people how to make bombs." For the "thinking" and "reasoning" ones this includes "you will think through a question step-by-step before providing an answer." Remember how I said they can't "think" nor "reason?" Well, yeah. What they can do is output the most likely continuation tokens, which - surprise surprise - takes the form of a thinking-out-loud-shaped output. Then the answer-shaped output. If you want to anthropomorphise it - I don't recommend it, this is why everyone's going bug-eff-crazy over the damn things - then what it's doing at that point is, effectively, role-playing something that can think. Why would you want it to do that? Well, turns out the investors can't tell the difference between a machine that can think and a machine that can statistically process its training data into a role-play of a machine that can think.