cr4zy

What are you all using instead of Upwork?

For compression and long-running agents, may I suggest https://memtree.dev. We offer a simple API that compresses messages asynchronously for instant responses and small context leading to much higher quality generations. We're about to release a dashboard that will show you what each compressed request looked like, the token distribution between system, memory, and tool messages, along with memory retrievals, etc... Is this the type of thing that you're looking for?

For code it's actually quite good so far IME. Not quite as good as Gemini 2.5 Pro but much faster. I've integrated it into polychat.co if you want to try it out and compare with other models. I usually ask 2 to 5 models the same question there to reduce the model overload anxiety.

I discuss how the automation wave is already starting with white-collar job openings at a 12 year low in the U.S. I also talk about how we cannot simply count on taxing AI to support automated workers, as countries that don't tax will outcompete those who do. We therefore need international cooperation, based in MAIM, from the original Superintelligence Strategy paper.

I also discuss how bioweapons cannot be avoided via restricting open weight models as originally suggested in Dan's paper. Rather we need to heavily invest in bioweapon defense, and in particular use AI for wastewater monitoring and accelerating metagenomics (detangling mixed DNA).

I've added a "Thinking...." which sends server side events to keep the cxn open. Would love to hear if o1 models now work for anyone who they were broken for.

Open WebUI does offer an API, but I have it disabled for PolyChat.

One tradeoff of bringing your own API keys is that as you add more model providers, you get more billing accounts to deal with. Chorus also doesn't have an incentive to efficiently use your tokens. We save 67% on Anthropic token costs using Claude Caching. We also use cheaper "task models" for conversation titles, tagging, and parts of the RAG pipeline which all drastically cuts token costs.

For local models I highly recommend https://github.com/crizCraig/open-webui

They do the side by side thing that Chorus does and you can serve it to anywhere including your phone.

So it looks like some folks are getting errors with the non-streaming models, i.e. the o1 models. I think their long running cxns with zero packets may cause some networks to drop the requests. Will look into a hearbeat/keepalive on those.

It's not unlimited free unfortunately. After some free use, we provide monthly usage-tier plans. But there are no rate limits like other providers as you can move up to the next usage-tier.

Background chats are new in v5 of Open WebUI, so you can use it too. Overview has been there also, but it's kind hidden in the hamburger menu.

The upgrade/logout issue you're facing is likely due to not setting WEBUI_SECRET_KEY outside of your docker container. This causes all previous cookies to be unreadable as a new key will get generated by start.sh and won't decrypt the old cookies.

Cool! I should say most of PolyChat is open source at https://github.com/open-webui - just the combo models and payment are closed source right now. Open to arguments on making PolyChat fully open source as well!

Thanks! I definitely recommend launching quickly and iterating.

edit: I should add, I'm making only $30/mo in subs, but I just launched so we'll see!

Wow, this is awesome! Thanks for building. I didn't realize there was a protocol for streaming while rendering, though I noticed sumo.ai doing something similar for audio. Gemini with grounding is new to me also, very nice!

Another great session of George and John recording Oh My Love [1] shows what I think of as embracing rough but promising ideas along with balanities like tuning guitars, learning chords, etc... The Beatles engaged in this process much more than most musicians, despite it meaning most the stuff they played sounds relatively bad compared to playing tunes they'd practiced a bunch before. Is this due to work ethic or some love of creation? Idk, but it is fascinating to see the sausage get made.

[1] https://www.youtube.com/watch?v=yksV7YVuqdg

Allocating 20% to safety would not be enough if safety and capability aren't aligned. I.e. without saying Bostrom's orthogonality thesis is mostly wrong. However, I believe they may be sufficiently aligned in the long term for 20% to work [1]. The biggest threat imo is that more resources are devoted to AIs with military or monetary-based objectives that are focused on shorter-term capability and power. In this case, capability and safety are not aligned and we race to the bottom. Hopefully global coordination and this effort to achieve superalignment in four years will avoid that.

[1] https://drive.google.com/file/d/1rdG5QCTqSXNaJZrYMxO9x2ChsPB...

The nice thing about box breathing is that, unlike pursed lips for example, you can do it without anyone noticing, by quietly breathing through your nose. So if you're in a stressful meeting, you can calm yourself without anyone noticing :)

Turchin's political stability indicator predicts this due to rising economic inequality, overproduction of grads with advanced degrees, and exploding public debt [0][1].

However, civil wars usually occur in poorer countries [1][2]. Also the proportion of the world in civil war is 1% and falling exponentially [2].

Finally, Metaculus gives a 5% chance of US civil war before 2031, where a recent July bump is discussed in the comments (Roe, Jan 6) [3].

[0] https://www.nature.com/articles/463608a

[1] https://www.buzzfeednews.com/article/peteraldhous/political-...

[2] https://www.nber.org/reporter/2011number3/economic-shocks-we...

[3] https://www.metaculus.com/questions/6179/second-us-civil-war...

Notes: The Harvard December 2021 youth poll participants gave as 35% chance in their lifetime: https://iop.harvard.edu/youth-poll/fall-2021-harvard-youth-p...

Rasmussen 2018 poll: 33% likely, 10% very likely before 2023 https://www.newsweek.com/second-civil-war-likely-one-third-a...

Does anyone have a graph of this probabilty over time from a consistent source like rasmussen? https://www.rasmussenreports.com/public_content/politics/que...

With getpy[0], a python wrapper, you can get 200x faster map reads in parallel

    In [1]: import numpy as np
       ...: import getpy as gp
    
    In [2]: key_type = np.dtype('u8')
       ...: value_type = np.dtype('u8')
    
    In [3]: keys = np.random.randint(1, 1000, size=10**2, dtype=key_type)
       ...: values = np.random.randint(1, 1000, size=10**2, dtype=value_type)
       ...: 
       ...: gp_dict = gp.Dict(key_type, value_type, default_value=42)
       ...: gp_dict[keys] = values
       ...: 
       ...: random_keys = np.random.randint(1, 1000, size=500, dtype=key_type)
    
    In [4]: %timeit random_values = gp_dict[random_keys]
    2.19 µs ± 11.6 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)
    
    In [7]: %timeit [gp_dict[k] for k in random_keys]
    491 µs ± 3.51 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

[0] https://github.com/atom-moyer/getpy

Prediction machines, yet critically, taking actions that pursue states which we are not quite able to predict. https://reader.elsevier.com/reader/sd/pii/S0896627315007679?...

Trillion parameter networks are mentioned a few times, but Tesla is deploying much smaller networks than that (like tens of millions IMU). Trillion param networks are mostly transformers like GPT-3 (actually 175B) etc... that are particularly heavy vs Conv as they have no weight sharing. Tesla is definitely starting to use transformers though, e.g. for camera fusion and evidenced by their focus on matrix multiply in dojo asic's vs the conv asics they have in the on-vehicle chips.

Story was deleted, here's the archive link: https://web.archive.org/web/20210101220230/https://ex-apple-...

Make sure to disable JavaScript.

Found this through a d3 tutorial on how to create these types of visualizations: https://observablehq.com/@mbostock/the-impact-of-vaccines

Their version control integration is the best git UX imo. Especially love change lists for organizing individual lines of code into commits.

Interesting that Singapore also is the 4th least corrupt country per the Corruption Perceptions Index - one of the biggest discrepancies with the Democracy Index.

This is very much needed. I started to create something similar a few years back in case it's helpful. https://github.com/crizCraig/benjamin/blob/master/benjamin.p... Mostly it dealt with recommending reservations.

Here's my boilerplate bash header with comments describing each option, which I find practically helpful for starting new scripts

https://gist.github.com/crizCraig/f42bc250754bed764ada5f95d1...

raw: https://gist.githubusercontent.com/crizCraig/f42bc250754bed7...

This problem of run-away excitation in wetware reminds me of exploding gradients in artificial neural nets. We try to handle this with data normalization, batch normalization, and gradient clipping of different sorts (although unless the clipping is incorporated into the loss as in PPO (Schulman), it's very brittle and dependent on the data and network architecture.). So I wonder if we can glean something from these inhibitory neurons for artificial nets. The opposite problem of vanishing gradients results from too much inhibition, which happens in recurrent neural nets - so it's definitely a balance. Currently PPO does the best job IMO, but is specific to reinforcement learning.

The backend for this project actually has a wider scope - https://github.com/botleague/botleague, with our self-driving sim being the first use-case for it. Everything is open source and open data - sans some temporary keys used to communicate evaluation data between problem providers and our score keeper server (https://github.com/botleague/botleague-liaison). The hope is that others will post problems as well - of which combinations can be solved by more and more general and sophisticated bots. Curious as to what you all think of this self-driving leaderboard and the broader AI league idea. Very early stage here, but excited to get it out for y'all to try.

Highly recommend zero water or another in-house filter. In Mountain View, CA or South San Francisco - one zero water filter last me hundreds of gallons. In Mesa, AZ it only last me around 50 gallons but that is still competitive with a water and ice except that you don't have to drive the water back to your house.

Similar question on Quora from 2012[1] per sense:

  Vision: 10Mb/s => 100Mb/s
  Hearing: 30Mb/s
  Touch: 135Mb/s
  Smell: 100k neurons
  Taste: 100kb/s
  Proprioception: ??
  Balance: ??

  Total: ~10Mb/s=>~1Gb/s

Internal brain bandwidth is also worth mentioning as this is the last remaining wetware advantage over hardware due to the three dimensionally fully connected heterogeneous cortical substrate. I can't seem to find a figure on that though.

https://www.quora.com/How-much-bandwidth-does-each-human-sen...