mangoman

  This release unifies those capabilities with a Mixture-of-Transformers (MoT) architecture built around two towers. 
  Reasoner tower: A vision-language model (VLM) ... This serves as the ‘brain’ that reasons about the world before any generation happens.
  Generator tower: Generates future observations and action sequences. This tower uses a diffusion-based process to generate physics-aware video and action outputs that are conditioned on the reasoner tower’s understanding.

But based on my understanding of the Bitter Lesson (http://www.incompleteideas.net/IncIdeas/BitterLesson.html), this is precisely the wrong approach in the long term. I'm linking the actual text of the bitter lesson because I think it's misunderstood (or I just don't agree with how i've seen it used in discourse). Specifically:

  The bitter lesson is based on the historical observations that 1) AI researchers have often tried to build knowledge into their agents, 2) this always helps in the short term, and is personally satisfying to the researcher, but 3) in the long run it plateaus and even inhibits further progress, and 4) breakthrough progress eventually arrives by an opposing approach based on scaling computation by search and learning. The eventual success is tinged with bitterness, and often incompletely digested, because it is success over a favored, human-centric approach. 

I dunno, I thought that too for a while too, but there are a lot of new ideas in terms of architecture that may warrant massive training runs. Mamba and state space models are pretty interesting, but haven’t had their transformer moment yet because I haven’t really seen anyone go for broke on training it with a huge data set and model size. Even some of the more fundamental changes too like Kolmogorov–Arnold Networks or some of the ideas behind continuous back propagation haven’t really had the opportunity to be pushed to the limit. I think it’s still early days on what these models can do. And I say this as someone who bought a Mac m3 max 128gb ram, based on the hope that the on device training and inference work would eventually move locally. It’s encouraging to see the progress though and I hope it does move locally though.

There’s something off-putting about making a blog post about some splashy tech that’s is a fork of an open source project, and that tech not also being open source? It reads to me like “Hey, we thought the open source goose project was just okay, so we forked it to do it better. But we’re not going to contribute it back to and instead rename it.”

I think it probably wouldn’t be as weird if the project were a meaningfully different fork of it, but it sounds like it’s trying to accomplish the same goals as the open source project which I feel should probably be ported back? and renaming it seems sorta ungrateful? Kinda like that “you made this? I made this” meme. Maybe I just don’t have an understanding of how different the projects are though…

I guess what’s wrong with it? Let’s say it has read only access, new messages and calendar invites need approval. I’m not sure I understand the harm? I suppose data exfiltration, but like you could start with an allowlist approach. So the first few uses and reads take a while with allowing the ai to read stuff , but it doesn’t seem that crazy given it’s what we basically do with ai coding tools?

I’ve never built something like ICEBlock that puts me personally in the crosshairs of not just normal hacking attempts, but also the political will of the federal government. I can’t imagine the cess pool that is Joshua’s DMs. I think OP makes all the right assessments when examining how seriously ICEBlock is taking the risks here. The Android push notifications assertion is proof enough to make me raise a pretty big question, let alone the other issues raised.

Were I building something that I would want to assert the level of privacy claims that ICEBlock asserts, I would absolutely be taking any/all reports about security extremely seriously.

no that's incorrect - llama.cpp has support for providing a context free grammar while sampling and only samples tokens that would conform to the grammar, rather than sampling tokens that would violate the grammar

My bringing up it's age was mainly about it not being used in a LONG time, so using it now would seem like a hail mary. I made no mention about current day laws, so I'm not sure where your impression came from.

Right. this whole episode is about foreign policy, why not use the act that is meant for it?

...sigh... i mean i guess it really is just about saying they _can_ do it instead of actually _trying to do it_....

I don't think it requires the courts to agree - just that there's a burden or disadvantage and that it's in the "public interest" which seems like a pretty low bar to make up a story that sounds plausible. i think the idea that a trade deficit is a disadvantage is kinda brain dead, but it's plausible sounding enough to argue in court. throw in unequal tariff rates and it seems like an easier win than the IEEPA's emergency justification.

I'm not a lawyer or even close to it, but why wouldn't the trump admin use the tariff act of 1930? quote:

"Whenever the President shall find as a fact that any foreign country places any burden or disadvantage upon the commerce of the United States by any of the unequal impositions or discriminations aforesaid, he shall, when he finds that the public interest will be served thereby, by proclamation specify and declare such new or additional rate or rates of duty as he shall determine will offset such burden or disadvantage, not to exceed 50 per centum ad valorem or its equivalent, on any products of, or on articles imported in a vessel of, such foreign country"

it does cap it at 50%, but I mean it seems like a much easier way to justify the tariff. is there something else about it that isn't as practical (other than being almost 100 years old)

a paper with a similar idea on scaling test time reasoning, this is sorta how all the thinking models work under the hood. https://arxiv.org/abs/2501.19393

From the S1 paper:

Second, we develop budget forcing to control test-time compute by forcefully terminating the model's thinking process or lengthening it by appending "Wait" multiple times to the model's generation when it tries to end

I'm feeling proud of myself that I had the crux of the same idea almost 6 months ago before reasoning models came out (and a bit disappointed that I didn't take this idea further!). Basically during inference time, you have to choose the next token to sample. Usually people just try to sample the distribution using the same sampling rules at each step.... but you don't have to! you can selectively insert words into the the LLM's mouth based on what it said previously or what it wants to say, and decide "nah, say this instead". I wrote a library so that you could sample an LLM using llama.cpp in swift and you could write rules to sample tokens and force tokens into the sequence depending on what was sampled. https://github.com/prashanthsadasivan/LlamaKit/blob/main/Tes...

Here, I wrote a test that asks Phi-3 instruct "how are you" and it if it tried to say "as an AI I don't have feelings" or "I'm doing " I forced it to say "I'm doing poorly" and refuse to help since it was always so dang positive. It sorta worked, though the instruction tuned models REALLY want to help. But at the time I just didn't have a great use case for it - I had thought about a more conditional extension to llama.cpp's grammar sampling (you could imagine changing the grammar based on previously sampled text), or even just making it go down certain paths, but I just lost steam because I couldn't describe a killer use case for it.

This is that killer use case! forcing it to think more is such a great usecase for inserting ideas into the LLM's mouth, and I feel like there must be more to this idea to explore.

That would be missing the forest for the trees in my view. I could see it having an impact, but when 60% of people say that the country is headed in the wrong direction, putting up a candidate who was in power the last four years just isn’t going to work. Biden would not have won a primary, and neither would she have

on 2) are you referring to https://en.wikipedia.org/wiki/Dancing_plague_of_1518 ? I don't know very much of the history, but the veracity of the claims is specifically called out in the wikipedia page. Given the year, I'd wager that the truth is that they died of something else...

And that's a very strange reason to be skeptical of womens' description of their symptoms in a medical setting. Is there evidence that women are more prone to 'social contagion'? You yourself said that "women are more prone to auto immune diseases" and that they are known to be triggered by viruses...To me, skepticism (and in my view, cynicism) of patients is one of the major contributors to distrust of evidence based medical advice.

Taking your last anecdote as an example, just because the patient's self diagnosis is likely incorrect, that does not mean that the symptoms they're experiencing are all "made up" and should be met with skepticism. There may be other reasons they're experiencing symptoms. A doctor shouldn't just say "Because you thought it was long COVID even though you weren't diagnosed with COVID, I'm convinced you're making the whole thing up". That's just lazy and unsound.

This is pretty neat! I’m on mobile right now, but you mentioned that it’s VR ready - does the landing page work with WebXR? I’d love to try it out on my meta quest 3

Update, this is extremely well written satire. I’m laughing at myself for believing it but it also sounds so real.

Not sure if this is satire or not… couldn’t find other sources yet but it’s relatively new

Author here, happy to answer any questions as best as I can! A fun project that I hope will help others get models running on mobile devices.

I recently had an unusual health event that resulted in me passing out. My wife, who is a physician, thought it might be hypoglycemia, since i'm at high risk for diabetes. She found a super friendly endocrinologist who put me on a CGM for two weeks. I never hit the hypoglycemia range during those two weeks, so it didn't really explain what my issue... but honestly the data was SUPER interesting. Just observing the various spikes made me make healthier choices, or noticing when I was feeling extra tired and seeing if that correlated to not having eaten for little while, or eating something sugary before.

It's sort of like tracking your steps when you first get a smart watch. It may not have been the reason you got the device, but seeing the data, people are encouraged to act on it, even if you don't have an acute issue. since I didn't have a prescription, I couldn't get one here (didn't want to go through some sketch online site). I tried to get one from my family in India, but the prices were really high and they couldn't get the fancier one that tracks straight to your phone, so I didn't get one.

I think this could be a god send for preventing pre-diabetic people who would take preventative steps if it weren't such a pain in the ass to measure consistently.

I understand the sadness around not understanding it, it's fucking hard. however, there are more and more resources online getting published for how to get started understanding it that help with understanding the math at an abstraction that helps with learning how to build with it.

I would strongly strongly strongly recommend starting with karpathy's from 0 to hero neural networks youtube course - it starts with building a tensor library and back propagation, explaining it in a way that finally clicked for me https://www.youtube.com/playlist?list=PLAqhIrjkxbuWI23v9cThs...

jeremy howard also has a fantastic video that is more around how to use LLMs and such called a hacker's guide to language models - https://www.youtube.com/watch?v=jkrNMKz9pWU&t=607s

as i've dove more and more into it, i would strongly recommend trying to run things on your local machine too (llama.cpp, ollama, LM studio). that has helped me fight that feeling of like "are we all just going to be open AI developers in the end" and made me feel like you _can_ integrate these things into stuff you build by your self. I can't imagine how fucked we'd all be if llama was never opensourced. being old does not mean that you can't continue to grow, and remember that it's okay to feel overwhelmed about all this - many people are.

prashanth.world

I haven’t updated it in a long time, but hopefully going to be making some more regular updates soon. Truthfully I have a tough time sharing writing or thoughts in public (including hn comments) because I often find my opinions changing and posting in public is so static. i hope to overcome that mental hurdle and start writing again

FullStory | Staff Software Engineer | Full-Time | Remote (US timezones)

My name is prashanth, I'm the manager of the Mobile Enablement team here at FullStory. My team is currently hiring for a Staff Software Engineer, primarily focused on backend skills, but who is comfortable jumping in to Android Studio, XCode, and TypeScript if necessary.

What we're looking for

* Focused on Backend - Our backend is written in Go, using gRPC, GCP, Grafana, BigQuery for logs, BigTable / datastore for storage, GCS for more storage.

* Comfortable jumping into other stacks - We do touch Java, Objective-C, Swift, TypeScript and Rust in some of the work we do

* Staff level impact - can build abstractions, and evolve them over time.

* Understands Mobile Apps and Websites - knows what an Android View is, UIKit View. Also knows how Web apps work and can understand the similarities and differences between apps and websites.

I started as an engineer on our SDK focused team working on deep internals of Android for a couple of years, but as our Mobile product started to get more customers, it became clear that the way we process Mobile sessions and find interesting insights wasn't evolving. Our team is focused on helping teams build features for our Mobile product without having to understand the nitty gritty details of our SDK and mobile specific data structures.

If you're not familiar with FS, here's my take on the elevator pitch:

FullStory builds the best product analytics + session replay tools to help you understand what is frustrating your users and preventing them from using your Mobile App or Website. FullStory automatically capture what your users saw, what they interacted with, and all the information you'd need to power your product analytics, and lets watch how your users are using your app or website in a video-like experience, all with best in class privacy preserving features. With FullStory, You can delete all your manually instrumented Analytics.event("Login Clicked"), and use our APIs to instrument only the events and data that matter to your app or site

I've worked here for 3.5 years, and I still think there's so many interesting problems to solve and tools to build - here's a little about the history and what excites me about our future.

For websites, the problem is a lot simpler - there are APIs to listen for changes to the HTML structure. There’s nothing really close to that on mobile that provides any sort of structured vector representation of the UI and View Tree Hierarchy. Most of the competitors in this space are all either based on Bitmaps, screenshots, or don’t provide any sort of a session replay experience (focusing on the analytics only case). Ours is based on different paradigm. We can selectively capture the colors, shapes, text, etc. that a View would draw to a canvas. Why is this better? If a view contains sensitive information, we can still capture the shapes, colors, paths of the view, but omit the actual text or bitmap content of that view.

This role is particularly exciting because we’re just scratching the surface with what the higher fidelity our SDK captures can do. Imagine automatically surfacing when your users are spending a long time looking at ProgressBars or UIActivityIndicatorViews and how that impacts your conversions without needing to log any events, metrics or write code. Our team will have an outsized impact on the success of our mobile product, and the person who fills this role will strongly influence that impact and build a lot of the the fundamental building blocks that power those features.

If this sounds interesting, apply here: https://www.fullstory.com/careers/jobs/56457178-344f-47c7-98...

Thanks for your perspective! sounds like for inbound only, I don't really have very many concerns then.

on being able to do easy updates - yeah the biggest annoyance would be updating.

Do you have a recommended email server to run?

Yeah I've seen that before, the "only certain domains for emails" restriction, but I suppose I'm crossing my fingers that that wouldn't stop me.

Do you have a recommendation for an email server program to use?

Thanks for responding! I'm not quite sure I follow what the value proposition (over a generic analytics solution) would be then if the SDK doesn't know how to automatically detect the 'interesting' events and I'd have to add them manually, unless I'm missing something? Why is 'figuring out how to' track IoT specific events tricky? And why would 'your documentation' be enough of a differentiator? It seems like the important events would largely be based on various APIs provided Android/iOS, no?

I guess, I'm not really understanding how kraftful would save users time. And those other product analytic services have tons of integrations, AND they could also use those solutions for their landing pages / logged in web experience too. Imagine creating a 'funnel' like

Home screen (referrer=HN) -> product purchased -> app installed -> device connected

where the funnel tracks the funnel of users from HN all the way down to device connection. Does that make sense?

I realize that my concerns are a little pessimistic, sorry about that! I do think that there's something there, but I'm not quite sure I understand the overall value prop. I also noticed your site has something like a no-code 'white label' mobile app product for IOT device manufacturers, that seems really interesting! how does that tie in with the analytics piece?

I'm confused, how would this be different than an IOT app developer adding mixpanel, amplitude, etc and tagging the various events in their app? does the sdk somehow automatically figure out when the device that the user is trying to connect actually connects properly?

I generally agree that MVP tends to be cargo-culted rather than chosen as an intentional strategy. Most people who tend to spew the term often don't even know what they mean. I also agree on people tending to focus way too much on the 'minimum' and not nearly enough on the 'viable'. for example, how can you know if investing in the UI or design of a product is necessary for viability? I've seen that any interesting UI is often cut in the effort to find an MVP... but, well, sometimes an interesting design actually increases viability.

test

The linked post in TFA (which is essentially the OP's main argument in favor of web3) describes how the PC won, even though it was deficient in almost every dimension (worse CPU, worse memory, storage, etc) than existing mainframes, except for one dimension: price.

It then goes on to describe how blockchain is essentially the same thing - it's a "worse" database in every way, except one: it's not controlled by a single entity

The problem with this analogy is that price is a very very different and meaningful dimension than something like memory, or storage. making something cheaper is different than making something faster. Making something faster doesn't make it more accessible. The PC was more accessible than any other computer out there.

To compare price to "being controlled by a single entity" is totally wrong. Sure, the data is "accessible" in that it's public, but to use the technology is actually way more expensive than a traditional database. in fact, calling blockchain better while ignoring the price of storing data in the blockchain is like, exactly the opposite reason why the PC won!

I'm actually pro making data way more accessible, and I wonder if there's a web3 app out there that isn't some sort of financial instrument, and not about hoarding tokens in the hope that the price for them goes up, but actually about distributing data ownership in a way that people who didn't "get in early" can actually have a seat at the table.