I lived my whole life in the twin cities and have a lot of friends, US citizens, who are too scared to go out to eat right now because of the ICE raids. If that wasn’t the point it is certainly the effect. I applaud Walz and Frey, and I will be ranking Frey first next time he’s up for reelection. Something tells me though he will be on to bigger things than mayor of Minneapolis.
The allegations of fraud made by the people invading my home and terrorizing my friends? I’ll take those with a grain of salt.
And for the record, I’m not afraid of ICE, never said I was. ICE is racially profiling people, arresting them without cause, and deporting them without due process. I happen to be white, so that doesn’t apply in my case. It’s also coincidentally the same reason I feel safe posting a comment like this online. Free speech is being chilled in communities that ice targets, and I feel responsibility to relay what I’m hearing.
ICE also just gunned down a US citizen in the street, and that should scare everyone.
I have a hard time to understand that. The woman just accelerated from 0mph to what, 4mph? The officer was aware of the car. The car was taking a turn away from the officer.
That was no life threatening situation. If the officer had not shot, what would have happened? The woman would have left the scene while the officers already had the license plate information. So I am not sure what your take is here.
Also Minneapolis emergency services reports indicate that Jonathan Ross was not taken anywhere, but instead remained at the scene and then left with the feds.
Yeah, it was filmed from three different angles. She was told to drive, she drove into him, he ended up with internal bleeding. You can see the moment she looks right at him and hits the gas.
It is absolutely frightening that you lot can watch something so physically simple and somehow not understand what happened.
What you write is fully compatible with what I have written.
So you think the officer defended his life in this situation or that he thought he was? I don't think so and I think it's clearly visible that's not the case.
> ICE shot someone who wasn't complying with several orders to stop her vehicle, to open her door,
The first one ordered her to leave, the next one told her to get out of her vehicle and reached into her vehicle without a warrent or justification.
Neither had any legal grounds for doing so, at least according to a civil rights lawyer who defended MAGA people deplatformed during the Biden administration.
Traffic enforcement is not within ICE's remit, particularly in this case when wide angle footage with long time frames show traffic passing both cars that are across a single lane.
> Now hitting the gas pedal and hitting an agent with your big fat SUV:
Perhaps avoid the false hyperbole.
> It's basically the same story: the left using anything it can to create martyrs
The "martyrs" are being created by the incompentance and over reach of ICE.
I don’t understand why you’re bringing up Obama. Nobody is defending him, and I think you might be inferring some sort of tribalistic defensiveness without any evidence. I see that sort of thing a lot, and I’ve never understood it.
Please correct me if I misunderstood the point you were trying to make.
The point I'm trying to make is that highly respected and even lauded individuals kill American citizens and people aren't generally afraid of them, so the fact that "they've shot and killed american citizens" on its own doesn't explain any fear. But I had misread the original question so my point was not quite relevant because the question was stated with justified fear already implied.
Currently going on? You mean the ones which were prosecuted a few years ago? Or are you talking about that "journalist"'s YouTube video that was hyped by the government to cause misinformed rage?
Yes I am talking about that "journalist"'s YouTube video, like I said the evidence out there is accessible on the Web. If there is a nanometer evidence that implies your tax money goes to Somalians bank account, you need to investigate that.
I'm pretty traditional conservative and I don't feel that our tax money should be funding child care directly like it does these days. I want to reduce the debt and taxes.
Looking at the "journalist", I really don't understand what people say he uncovered. He showed up with a bunch of people and expected to enter a non-public facility with children to investigate and when he was denied entry, made huge claims about how over half of the state's funding is going to illegal
Immigrants. I saw zero proof of anything in their "reporting" nor did I see anything credible in their approach. I fully support investigative reporting, but there's a bar that needs to be meet to be considered a journalist and rage content creators do not meet that bar.
If you want to link to specific videos that you feel do meet this bar, I'm happy to watch them, but as of now, I do not believe their reporting one bit.
This same guy, Nick Shirley, traveled to Ukraine and made a video about how there really wasn't a war and the Ukranians were just spending our money on luxury cars. A reporter, Caolan Robertson, who has spent years covering the war got an interview with him and called him out on his bullshit. If he lied about Ukraine, I am very disinclined to trust him about Somalians.
I'm not sure how this relates to what I said. The links you gave were to fraud found in Brooklyn. Did Nick Shirley uncover these cases?
I do believe that there's some fraud. I even believe there's a substantial dollar value of fraud going on. I do not believe that they uncovered 9 billion dollars in fraud visiting 5 Somali run daycare facilities and being denied entry into them with a camera crew and armed security.
Well, it seemed to work well enough on you. Maybe try and think harder about your information sources. They seem to be lying to you, and you don't seem to be able to tell.
I'm way more concerned about my tax money going to Donald Trump and his family and cronies. Followed by it going to billionaires and corporations. You know things that are actually happening today. I'm not very concerned at all by fluffed up recycled true crime by d tier content creators that the administration has turned into propaganda and people like you are pushing at every opportunity.
Seriously. What percentage of my taxes are going to a "Somalians bank account" as opposed to the greedy, blatantly corrupt pigs in power? I’m guessing there’s an order of magnitude difference, if not several. They are draining the coffers and laughing in our faces about it.
Your question is essentially equivalent to "Why are you scared of the Gestapo if you aren't a Jew?"
It's not a coincidence that a winner of the Nobel Prize gifted his medal to the then Propaganda Minister Joseph Goebbels[1], exactly as María Corina Machado gifted her Nobel medal to Trump. History doesn't repeat itself, but it sure does rhyme. If you close your eyes to all these similarities, you're going to be in for a big surprise.
> Obama deported 3 million immigrants with ICE while he is president
Obama did it the right way, didn't he? Obama's administration managed deportations effectively and humanely, prioritizing criminals and recent arrivals through programs like Secure Communities for over 3 million removals while upholding due process, minimizing community disruptions, and avoiding widespread violence or errors like wrong-country deportations.
> The woman who got shot thought she was on a movie set, tried to act cool and then attempted to run over an officer.
Citation needed. Absolutely wild to claim to know the thoughts of a dead woman. Do you speak for the dead also?
Shills keep bring up Obama deportations. Obama's administration primarily followed the law, observed due process and habeas corpus, didn't send people to concentration camps in other countries, and didn't shoot mother's in the head and then call them terrorists.
It should be clear why the response is so different, there's no need to act naive
1. US Police is very well known to act on emotions instead of actual law or protocol
2. Killing her didn't make the officer safer, now he was being approached by an effectively driverless vehicle.
3. If you put yourself infront of the car for fuck all reason that is your fault, especially as a "LEO" you should know better, but I guess that is what you get when you employ power hungry people with 0 training.
For what it's worth, the ICE agent who shot Good, Jonathan Ross, worked for US Border Patrol for eight years and has been working for ICE for a decade. Further, "Ross testified in December that he was 'a firearms instructor, an active shooter instructor ... a field intelligence officer, and ... a member of the SWAT team, the St. Paul Special Response Team'."[0]
And yet he put himself in front of a vehicle that had a driver and a running engine, and partnered up with other agents who acted as a group with zero cohesion, issuing conflicting instructions, escalating the tension of a traffic infringement they had no actual legal authority to engage with.
Every trained professional I've communicated with in regard to this incident has effectively shaken their head and referred to it as a clown show of epic proportions, a textbook example of how not to engage with the public, an example of how authoritarian states deal with people they have no regard for.
Let's be honest, a great many US enforcement types come to firearms use with an any excuse approach coupled with an absence of ability to de-escalate situations. They act like walking cans of petrol looking for a tinder to throw themselves on.
Just to be clear, I think the situation is disgusting, entirely unprofessional, and intentionally violent. Maybe the shooter didn't make those mistakes to have a reason for murder, but the idea that someone of his experience would make such a completely foolish mistake is absurd.
Nobody, as far as I know, has any intention of hitting me with their car. Yet, for about a dozen simple reasons, if I'm crossing a walkway, driveway, or whatever with a car also trying to enter the road by driving through the area I'm walking, I often will walk behind the vehicle. People make mistakes. There are blind spots. Maybe they're having an emergency. Maybe they're intoxicated. Or maybe they do want to hit someone with their car.
It's just absurd to think this was appropriate behavior by a seasoned professional. It's not even the appropriate behavior for a reasonably developed child.
> The woman who got shot thought she was on a movie set, tried to act cool and then attempted to run over an officer.
Your mind-reading abilities are malfunctioning. There's clear video evidence which disproves your claim. That raises serious questions about your good faith in this discussion.
> keeping the United States safe
From who exactly? Perhaps we could have a foreign country perform an operation to remove the people directing ICE to behave like an occupying force attempting to frighten US citizens into submission. That would help keep the country safe for democracy.
Do you think democracy in the US is worth preserving? Because it sure doesn't seem that way.
> For the other comment, you can check yourself if you don't believe in fraud.
This is a completely irrelevant distraction that just adds to the impression of bad faith from you.
Humans don’t learn to write messy complex code. Messy, complex code is the default, writing clean code takes skill.
You’re assuming the LLM produces extra complexity because it’s mimicking human code. I think it’s more likely that LLMs output complex code because it requires less thought and planning, and LLMs are still bad at planning.
Totally agree with the first observation. The default human state seems to be confusion. I've seen this over and over in junior coders.
It's often very creative how junior devs approach problems. It's like they don't fully understand what they're doing and the code itself is part of the exploration and brainstorming process trying to find the solution as they write... Very different from how senior engineers approach coding when it's like you don't even write your first line until you have a clear high level picture of all the parts and how they will fit together.
About the second point, I've been under the impression that because LLMs are trained on average code, they infer that the bugs and architectural flaws are desirable... So if it sees your code is poorly architected, it will generate more of that poorly architected code on top. If it sees hacks in your codebase, it will assume hacks are OK and give you more hacks.
When I use an LLM on a poorly written codebase, it does very poorly and it's hard to solve any problem or implement any feature and it keeps trying to come up with nasty hacks... Very frustrating trial and error process; eats up so many tokens.
But when I use the same LLM on one of my carefully architected side projects, it usually works extremely well, never tries to hack around a problem. It's like having good code lets you tap into a different part of its training set. It's not just because your architecture is easier to build on top, but also it follows existing coding conventions better and always addresses root causes, no hacks. Its code style looks more like that of a senior dev. You need to keep the feature requests specific and short though.
> About the second point, I've been under the impression that because LLMs are trained on average code, they infer that the bugs and architectural flaws are desirable
This is really only true about base models that haven’t undergone post training. The big difference between ChatGPT and GPT3 was OpenAI’s instruct fine tuning. Out of the box, language models behave how you describe. Ask them a question and half the time they generate a list of questions instead of an answer. The primary goal of post training is to coerce the model into a state in which it’s more likely to output things as if it were a helpful assistant. The simplest version is text at the start of your context window like: “the following is code was written by a meticulous senior engineer”. After a prompt like that the most likely next tokens will never be the models imitation of a sloppy code. Instruct fine tuning does the same thing but as permanent modifications to the weights of the model.
Image generation and image input are two totally different things. This is about feeding text into LLMs as images, it has nothing to do with image generation.
Yeah but IIUC they're both just representations of embeddings in a latent space, translated from one format to another. So if the image interpretation of a text embedding is full of hallucinations, it's unlikely that the other direction works well either (again, IIUC).
That said, I'll be interested to see what the DeepSeek model can do once they've trained it in the other direction. It'd be great to have it output architecture diagrams that actually correspond to what it says in the chat.
I don't know if this game was vibe coded, but it certainly could have been. Most notable thing about this post is probably that vibe coded games are good enough now to fool HN.
Fair enough, it is impossible to know. I'm making an assumption based on the fact that the use of AI wasn't mentioned anywhere I could see, and people usually mention it.
Same. There are really only two features I care about in a phone: a high refresh rates and weight. At 165 grams the iPhone air is by far the lightest 120hz phone apple has ever made. Second place is the iPhone 15 Pro at 187 grams. Getting ready to ditch my 15 pro.
Me too, and I’m planning the same upgrade. Always wanted to downsize but 60Hz was a deal breaker for me. Been using 120-144Hz+ displays exclusively since the VG248QE in 2013.
I was quite surprised to see this entire thread full of HN users who apparently want some brick phone to doom scroll lying flat on a table all day until the battery dies.
>> For each LLM, we extract static, token-level embeddings from its input embedding layer (the ‘E‘matrix). This choice aligns our analysis with the context-free nature of stimuli typical in human
categorization experiments, ensuring a comparable representational basis.
They're analyzing input embedding models, not LLMs. I'm not sure how the authors justify making claims about the inner workings of LLMs when they haven't actually computed a forward pass. The EMatrix is not an LLM, its a lookup table.
Just to highlight the ridiculousness of this research, no attention was computed! Not a single dot product between keys and queries. All of their conclusions are drawn from the output of an embedding lookup table.
The figure showing their alignment score correlated with model size is particularly egregious. Model size is meaningless when you never activate any model parameters. If Bert is outperforming Qwen and Gemma something is wrong with your methodology.
Note that the token embeddings are also trained, therefore their values do give some hints on how a model is organizing information.
They used token embeddings directly and not intermediate representations because the latter depend on the specific sentence that the model is processing. Data on human judgment was however collected without any context surrounding each word, thus using the token embeddings seem to be the most fair comparison.
Otherwise, what sentence(s) would you have used to compute the intermediate representations? And how would you make sure that the results aren't biased by these sentences?
You can process a single word through a transformer and get the corresponding intermediate representations.
Though it sounds odd there is no problem with it and it would indeed return the model's representation of that single word as seen by the model without any additional context.
Embedding models are not always trained with the rest of the model. That’s the whole idea behind VLLMs. First layer embeddings are so interchangeable you can literally feed in the output of other models using linear projection layers.
And like the other commenter said, you can absolutely feed single tokens through the model. Your point doesn’t make any sense though regardless. How about priming the model with “You’re a helpful assistant” just like everyone else does.
It’s mind blowing LeCun is listed as one of the authors.
I would expect model size to correlate with alignment score because usually model sizes correlate with hidden dimension. But also opposite can be true - bigger models might shift more basic token classification logic into layers and hence embedding alignment can go down. Regardless feels like pretty useless research…
So instead of next token prediction its next event prediction. At some point this just loops around and we're back to teaching models to predict the next token in the sequence.
I think it’s helpful to remember that language models are not producing tokens, they are producing a distribution of possible next tokens. Just because your sampler picks a sequence of tokens that contain incorrect reasoning doesn't mean a useful reasoning trace isn’t also contained within the latent space.
It’s a misconception that transformers reason in token space. Tokens don’t attend to other tokens. High dimensional latents attend to other high dimensional latents. The final layer of a decoder only transformer has full access to entire latent space of all previous latents, the same latents you can project into a distribution of next tokens.
> Just because your sampler picks a sequence of tokens that contain incorrect reasoning doesn't mean a useful reasoning trace isn’t also contained within the latent space.
That's essentially the core idea in Coconut[1][2], to keep the reasoning traces in a continuous space.
What the model is doing in latent space is auxilliary to anthropomorphic interpretations of the tokens, though. And if the latent reasoning matches a ground-truth procedure (A*), then we'd expect it to be projectable to semantic tokens, but it isn't. So it seems the model has learned an alternative method for solving these problems.
You’re thinking about this like the final layer of the model is all that exists. It’s highly likely reasoning is happening at a lower layer, in a different latent space that can’t natively be projected into logits.
That’s true yeah. The model can do that because calculating latents is independent of next token prediction. You do a forward pass for each token in your sequence without the final projection to logits.
So you're saying that the reasoning trace represents sequential connections between the full distribution rather than the sampled tokens from that distribution?
The lower dimensional logits are discarded, the original high dimensional latents are not.
But yeah, the LLM doesn’t even know the sampler exists. I used the last layer as an example, but it’s likely that reasoning traces exist in the latent space of every layer not just the final one, with the most complex reasoning concentrated in the middle layers.
I don't think that's accurate. The logits actually have high dimensionality, and they are intermediate outputs used to sample tokens. The latent representations contain contextual information and are also high-dimensional, but they serve a different role--they feed into the logits.
The dimensionality I suppose depends on the vocab size and your hidden dimension size, but that’s not really relevant. It’s a single linear projection to go from latents to logits.
Reasoning is definitely not happening in the linear projection to logits if that’s what you mean.
My personal theory is that it’s an emergent property of many attention heads working together. If each attention head is a bird, reasoning would be the movement of the flock.
Either I'm wildly misunderstanding or that can't possibly be true--if you sample at high temperature and it chooses a very-low probability token, it continues consistent with the chosen token, not with the more likely ones
Attention computes a weighted average of all previous latents. So yes, it’s a new token as input to the forward pass, but after it feeds through an attention head it contains a little bit of every previous latent.
Are you running both DearImGui visualisation and training locally? If not, how can one use it in the client-server mode? I think this is the most common requirement for visualisation libraries in Deep Learning.
The rendering is done with OpenGL, and for remote viewing I just render to an offscreen framebuffer and stream it back to the client with WebRTC. The code for that isn’t public yet, still needs some cleanup.
