readyplayeremma

MapLibre's globe mode is both fantastic and performant. Also, it's literally just the one option to change it, and your tile formats/CRS don't need to change either.

It's the easiest way to escape from web mercator projections with no real downsides that I have discovered yet. Also, there is a built-in control if you want to offer a button to toggle between web mercator view, and globe view, since it's all just rendering changes.

The models are right here, one of the first links in the post: https://huggingface.co/collections/Qwen/qwen3-omni

edit: Nevermind, in spite of them linking it at the top, they are the old models. Also, the HF demo is calling their API and not using HF for compute.

The article’s real contribution is in highlighting evidence of complex behavior in living systems that often get excluded from definitions of "intelligence". In doing so, it invites deeper philosophical reflection, even if it doesn’t mount that reflection itself.

The server does not generate the tokens, the client generates the tokens. The server is supposed to be able to verify that they were generated by a client who was granted the authority to generate them, but not which client did so. At least, not without side-channel information.

> The main building block of our construction is a verifiable oblivious pseudorandom function (VOPRF)

I am not sure how well tested that primitive is, but it definitely appears to be more than the server handing clients tokens and then pretending not to know who it gave them to.

The referenced paper: https://petsymposium.org/popets/2018/popets-2018-0026.pdf

You don’t have any specific routes to random internet addresses though. And Tailscale would not either. Unless your Windows server is running BGP, all your Internet traffic is hitting the default route.

You can just run whisper on the conversations as a background job populating the text versions of all the user inputs, so it doesn't interfere with the real-time latency.

Maybe it was before, but now the entire world knows it well. To change it would be taking an incredibly recognizable name—like Twitter for instance—and then just throwing it away. ChatGPT is forever part of history, and almost everyone recognizes it immediately at this point.

> B., a senior officer who used Lavender, echoed to +972 and Local Call that in the current war, officers were not required to independently review the AI system’s assessments, in order to save time and enable the mass production of human targets without hindrances.

> “Everything was statistical, everything was neat — it was very dry,” B. said. He noted that this lack of supervision was permitted despite internal checks showing that Lavender’s calculations were considered accurate only 90 percent of the time; in other words, it was known in advance that 10 percent of the human targets slated for assassination were not members of the Hamas military wing at all.```

So, there was no human sign-off. I guess the policy itself was ordered by someone, but all the ongoing targets that were selected for assassination were solely authorized by the AI system's predictions.

This sentence is horrifically dystopian... "in order to save time and enable the mass production of human targets without hindrances"

I would like more details on Gemini's 'native' multimodal approach before assuming it is something truly unique. Even if GPT-4V were aligning a pretrained image model and pretrained language model with a projection layer like PaLM-E/LLaVA/MiniGPT-4 (unconfirmed speculation, but likely), it's not as if they are not 'natively' training the composite system of projection-aligned models.

There is nothing in any of Google's claims that preclude the architecture being the same kind of composite system. Maybe with some additional blending in of multimodal training earlier in the process than has been published so far. And perhaps also unlike GPT-4V, they might have aligned a pretrained audio model to eliminate the need for a separate speech recognition layer and possibly solving for multi-speaker recognition by voice characteristics, but they didn't even demo that... Even this would not be groundbreaking though. ImageBind from Meta demonstrated the capacity align an audio model with an LLM in the same way images models have been aligned with LLMs. I would perhaps even argue that Google skipping the natural language intermediate step between LLM output and image generation is actually in support of the position that they may be using projection layers to create interfaces between these modalities. However, this direct image generation projection example was also a capability published by Meta with ImageBind.

What seems more likely, and not entirely unimpressive, is that they refined those existing techniques for building composite multimodal systems and created something that they plan to launch soon. However, they still have crucially not actually launched it here. Which puts them in a similar position to when GPT-4 was first announced with vision capabilities, but then did not offer them as a service for quite an extended time. Google has yet to ship it, and as a result fails to back up any of their interesting claims with evidence.

Most of Google's demos here are possible with a clever interface layer to GPT-4V + Whisper today. And while the demos 'feel' more natural, there is no claim being made that they are real-time demos, so we don't know how much practical improvement in the interface and user experience would actually be possible in their product when compared to what is possible with clever combinations of GPT-4V + Whisper today.

Really awesome project!

A technical nitpick though: Routes can be asymmetric—going across one path in one direction and another for the opposite. This means that your tool potentially doesn't show the route packets from the user took to reach your server, but rather the route packets took from your server to reach the user. I believe that querying with BGP looking glass tools would allow you to construct the route in either direction, but it is maybe a bit less cool looking than the real-time traceroute that is a result of actual traffic.

LLaVA-Plus maintains a skill repository that contains a wide range of vision and vision-language pre-trained models (tools), and is able to activate relevant tools, given users’ multimodal inputs, to compose their execution results on the fly to fulfill many real-world tasks.

> People casually asserting that software is equivalent to humanity will be a non-negligible thing to consider, as irritating and poorly-founded as it seems.

> If the reproduction isn't pixel-perfect, but merely obvious and overwhelming, how do you refute that philosophically to people who refuse a distinction between 50GB and a human life?

Software equivalence to humanity is a very philosophical question that many sci-fi writers have approached. But our primary issue related to this technology does not depend on anyone making a determination there.

The challenge is that losses to livelihood from this technology are going to come from far broader impacts than copyright alone. Copyright disputes are just the first things to get everyone's attention.

Let's say we err on the side of protection of copyright, and all training data must be fully licensed, in addition to users being responsible for ensuring outputs did not accidentally reproduce something similar to a copyrighted work, even if it was part of the licensed training dataset. Great! This fixes the problem of lost value for the owners of copyrights. Companies will face a slight delay and slightly increased costs as they license content; however, in the end, model capabilities will be the same and continue to increase.

The number of jobs that actually cannot be performed without humans will continue to dwindle — livelihoods will be lost at essentially the same scale despite upholding copyrights.

The only way we can handle a technology capable of reducing most need for human labor is by focusing on planning and executing a smooth transition toward an economy with more people than jobs — aiming for minimal human suffering during this process.

A mass loss of human jobs does not need to mean a mass loss of livelihood if our society is prepared to transition to a universal basic income. After all, human life is far more than just a job. We have the opportunity for much more fulfilling lives if we plan this transition well. We must understand that this is a far larger issue than copyright - copyright disputes are just one of the first symptoms of this disruptive process.

This is the best path forward I think. And it will become increasingly sensible as things continue to evolve. AI wasn't necessary to violate copyright before, and it isn't necessary today.

The determination of copyright violation should be made against the output of the model in the event that someone uses it for commercial purposes.

If the models have a risk of generating copyrighted content, it will be up to the consumers of the system to mitigate that risk through manual review or automated checks of the output.

The SGPT model is a very high performing text embeddings model adapted from a decoder. Using the same techniques with Llama-2 might perform better than you expect. I think someone will need to try these things before we know for certain. I believe there is still room for significant improvement with embedding models.

The GitHub project has no license file that I see. Does anyone know if this is going to be released under an OSS license of some kind?

I can also add one more data-point in favor of Elastic / OpenSearch. They benefit from a long history of providing search-specific features. Including the ability to write custom re-ranking functions to combine the benefits of traditional TF/IDF style search with the modern benefit of vector search techniques. And you can easily use OpenSearch with state of the art open embedding models like SGPT that use 2048-dimensional vectors. Plus, it is designed to be highly scalable and distributed.

Given how well OpenSearch works and scales, I would find it hard to justify a specialized vector-specific database unless it brought A LOT of new benefits to the table. And I am not currently aware how any of them would actually do that.

Also, OpenSearch provides all of that out-of-the-box. You just configure a vector field mapping and start inserting your data. No need for an add-on plugin/extension. It just works.

This is the correct viewpoint. ChatGPT is one specific implementation of the technology, which was most people's first exposure to it. The broader applications of the technology itself are still in the very early stages, but are already making significant impacts.

I can confirm this happens with the API as well though. But it does work if you ask it to output the quote in YAML format somehow.

acknowledgeFear: I must not fear.

understandEffect: Fear is the mind-killer.

describeImpact: Fear is the little-death that brings total obliteration.

faceFear: I will face my fear.

letItPass: I will permit it to pass over me and through me.

observeAbsence: And when it has gone past, I will turn the inner eye to see its path.

finalResult: Where the fear has gone there will be nothing. Only I will remain.

Except the person in the room also had to write the book by looking at examples of Chinese.

The search code appears to be hand-written though. You can find the commit history for the core search functions here: https://github.com/buddhist-uni/buddhist-uni.github.io/blame...

It's kind of neat that they did hand write it in a way. A fun project I'm sure. And they even made it run in a worker to keep things even more quick and responsive.

This is all pretty awesome. However, with regard to search, the zero latency is indeed fast, and this is because it is client-side entirely. It appears to be hand written for this site.

I think the quality of search recall and ranking could be improved though. It has zero tolerance for misspelling for instance, and cannot match partial words while you are typing the terms. Still, it is pretty cool. And open source too, which is even cooler.

This is a good overview of a number of emergent things that remain unexplained about the capabilities of some large language models:

https://www.scientificamerican.com/article/how-ai-knows-thin...

You can follow the multitude of things from there to the papers published that describe each of the cases in greater depth.

I understand that you may believe your ability to think is magic, but I assure you, you are also just a statistical predictor, and not magic. Thinking and cognition themselves, beyond the medium they happen within, are moving out of the realm of philosophy and into the realm of science now. It just so happens that quite a lot of people are uncomfortable with the idea of not being as special and unique in this way anymore.

What size model are you quantizing and comparing? The interesting thing about quantization, is how the larger the number of parameters, the less of a difference it makes to quantize the weights, even to an extreme degree when working with the largest parameter models. For small models is can be a disaster though.

Thank you for so eloquently conveying that. When a broader set of the world's population understands this, I hope for all our sake that we manage to cooperate effectively enough to address some of those realities you mention.

AI is not a well-defined term. This is more likely a specific machine learning technique that is designed to identify all the boring "normal stuff" and ignore it, and then do that at very large scale. By doing that, you can find the interesting parts so that humans with more limited resources can determine if those newly flagged things need to be added to the boring list, or if they do represent something truly interesting.

edit: After reading the article a bit more, it is using a random forest classifier. This is almost certainly not meeting the definition of what many here are thinking when the term "AI" is being used in the title. The term is clearly used here for marketing purposes.

Our existing computing platforms are very specific and very limited. Without knowledge of that specific design and the accompanying limitations, how could you pre-craft a non-interactive "exploit" that could be executed by such a system? I do think that it becomes more possible if we have AGI-like systems doing detection and analysis, but we do not. In addition, any universal exploit against AGI systems would probably have to be universal enough to also affect human intelligence.

It seems unlikely any alien malware could be targeted to affect our computing platforms. However, if they had preexisting knowledge of computing as implemented on Earth, it would be less complex. If there was a type of malware universal enough to do so, it would probably be some kind of attack against the universal notion of intelligence itself. In which case, you may need to isolate it from humans as well.

I think the most likely scenario for receiving any kind of alien Trojan horse signal, would be if the signal was some kind of instructions for how to create and execute an alien AI as encoded in the signal. However, that would require complex analysis and human intervention to build. Unless we reach a point where AGI systems can search, interpret, and implement instructions from said signals, any threats would require significant involvement from humans in order to materialize.

At least, those are my initial thoughts.

Human history is filled with examples of us thinking we are special or at the center of the universe, etc. For much of history, and even now, we don't even manage to equally consider all human life. Time and time again we discover how wrong we were. Most often it is not that we ascribe too much credit, or not enough credit to another possible form of life. Rather, we spend too much effort believing we and our conscious experience is the only kind that matters, and anything different from that cannot be sentient intelligence.

At some point we'll create intelligence more advanced than our own, and hopefully it has the capacity to be able to recognize that we can still have meaning even though we might seem like the ones with an uncertain level of intelligence at that point. If we want to help ensure that is the case, we should start to ascribe some more meaning and sanctity to the other forms of life besides ourselves. That next thing is probably going to take after its creator, and if we have no regard for any life but our own, we shouldn't expect it to learn something differently from us.

The day to day experience of our lives is no less magical if human consciousness or life is not uniquely special, or even if the universe we exist in was entirely simulated. We experience it the same in any case.

We don't have to be a uniquely special case. It's not necessary for us to be in order for life to have meaning or be sacred.

So let's not dismiss the possibility of consciousness or sentience in any other potential life forms around us, natural or artificial. We need to try and err on the side of caution, and reevaluate often as we learn more. And let us agree together that all potential intelligence or sentience should be handled with as much care and respect as we can reasonably provide.

That is really cool! What a fun way to visualize beamforming!