creativeSlumber

> then there are 10^100 possible paths through that session.

Is this correct? in practice you would only be deciding what the next turn is going to be. Because the turn after the next turn is determined by the outcome of the next turn. So shouldn't this number be 10 * 100?

What's different this time is that LLMs require orders of magnitude more memory than we could have ever used before. And I think it's getting obvious by now that inference will move to local compute, So everyone's laptops and phones would need a lot of memory.

is this just a fancy way of saying that sometimes humans makes quick heuristic based decisions, and sometimes they think through things thoroughly?

I don't think there's any problem category that is strictly a quick heuristic decision or something that you will think through thoroughly always. I think it's more about how much time you have. If you don't have time you'll make a quick heuristic based decision. If you have more time you will think through it more. Imagine you're driving and suddenly like a branch falls on the road right in front of you. And you need to avoid it. Your brain will quickly use a heuristic based approach to avoid the branch. But on the other hand, if this branch was already there on the road and you saw it from far away, you will probably take a lot more time to think through and figure out which path you need to take.

Wouldn't you need a classifier to even decide if it is system 1 or 2? How capable does this classier need to be?

How relevant is this fast/slow thinking thing with regards to current frontier models?

I know a large organization who's built their AI framework completely around this concept, and I feel that it's not really meaningful concept with the capabilities of current models.

I was wondering about this one too.To make things worse I also use speech to text and that introduces its own inaccurate transcriptions and typos. Are there any reliable research around this ?

> and I am using DeepSeek directly via the API.

Do you know if they retain your prompts or use it for training?

> And the water used must be recycled and self sustaining. If you need a million gallons fine but that’s all we can’t have data centers sucking up all the water.

They can have a closed loop cooling system, but it costs more energy to run and would eat into their profits. so they use evaporative cooling, cheaper, but consumes entire towns worth of water. If this is pumped from the ground, round water levels dry up in all around and peoples wells dry up.

its an ai written article it seems. probably written for engagement bait.

If the debt cancels out doesn't this mean that there was no debt ?

> 6. 200 OK, no answer Reasoning models sometimes put everything in the reasoning field and hand back content: null, finish_reason: "stop". 345 completion tokens, HTTP 200, nothing to show the user. A 200 tells you the request was served, not that there's an answer in it. No content and no tool call is a failure, throw and retry.

I ran into this so much I had to modify my harness to handle this.

do they lend money to their customers to buy their own chips?

> Do we use user feedback and de-identified data to improve ChatGPT and Codex in a holistic way? Yes.

This says that they trained on user sessions. The de-identification here I believe refers to removing PII, which doesn't matter here because the issue at hand is the content of the researcher's session where they likely discussed their approach tackling the Navier Stokes problem.

> Did any human or agent look at user data as part of the Navier Stokes effort? No.

If they trained the model on Lavent's chat sessions (PII removed or not), then this statement is meaningless as the model weights already contain that information.Given it's a new yet unreleased internal model, it is likely a 10+ trillion parameters (Astra is rumored to be 10 trillion), so the model can retain a lot more detail/info from training data.

Why is he leading the with the irrelevant part first ?

> And so does every LLM company.

Nope, not for enterprise users.No enterprise customers would use it if all of their internal business plans / trade secrets would end up in the model weights of the next OpenAI model. Imagine your competitor asking chatGPT a question and the model spitting out your business plan. These models can retain very specific fine grained data. I remember there were examples of them reproducing sections of their training data verbatim.

> Researchers had found signs of CKDu outside Central America, including in Sri Lanka, India, and Nepal. But it had never been documented in the United States.

I searched a bit more and found this interesting bit from an article talking about a similar epidemic in Sri Lanka.

> An analysis showed that trace metal ions such as calcium and magnesium that naturally occur in the hard water reacted with glyphosate. This meant that, instead of breaking down rapidly, glyphosate was able to persist in the environment for up to seven years in water and twenty-two years in soil.

https://www.thinkglobalhealth.org/article/mysterious-kidney-...

> Vibe: users are not opted out by default

> Vibe (Enterprise): customers are opted out of training by default

So they made it opt in for enterprise (how it should be), but intentionally made opt out for regular user.Basically saying "screw you: to regular users.

Any self respecting user should stop using them.

Agree, that it is an ad. Also, unless the full session (in a resume-able format, along all session artifact such as intermediate research/docs used to produce the final output) is also included in the commit, the session id is useless.

I presume this is Anthropic trying to their "usage" stats before their IPO.

their cache hit rate is 67%. In comparison the provider with the highest hit rate is at 95%.

this looks pretty cool. since it curves the the 3d map, what would it look like when you are driving through downtown highrises? would the tops of the buildings clash into each other?

what about steep uphill/downhill sections ?

This comparison is meaningless because China is the worlds factory (including a major portion of what's consumed in the US ). So that CO2 number is their populations usage plus what's required to manufacture good for the most of the world. A huge part of US CO2 consumption has been outsourced to China.

This is an interesting problem to tackle. It's not clear from the github readme what the output of this looks like, specifically what does it return to the LLM?

> The three axial flux motors are integrated per axle

I wonder why they need tree motors per axle.

> it's not important for a CEO to be good with software engineering

If you are the CEO of a company, you should have expertise in whatever your company does, and If your company is primarily a software company, then you should have expertise in software engineering. You cannot effectively manage something that you don't understand.

... that you pay to install on your machine.

I think you are missing the point here. what matters is for that user the local models are good enough for their use case.

this is one of the most popular options. Self hosted. https://immich.app/

> "An AI and a pair of human doctors were each given the same standard electronic health record to read"

This is handicapping the human doctors abilities. There is a lot more information a human doctor can gather even with a brief observation of the patient.

> did anyone on the team really not push back?

This is the real question. If they are serious about not doing something like this again, they NEED to look at what process failed and let something like this get proposed, designed, implemented and pushed to production. Usually things get reviewed at each stage. Did the people who pushed back on this get steam rolled? If no one pushed back, that's an even serious culture question and the entire org would need training.

A serious "we won't do it again", needs to be accompanied by a COE on this for identifying what went wrong, and identifying what guardrails can be put in place and then actually implementing them.

> Owning the decision

Owning a decision means you have something at stake if things go wrong. What would happen to Jack if this decision turns out to be wrong? Any consequences?

> I've worked at companies that are literally 10x more effective than other competitors in the market purely due to good engineering practices.

Most big tech companies get taken over by leadership with no tech background eventually and the engineering bar drops to the floor.

Are the model weights burned into the silicon / part of the architecture? Or can you update the model weights on these chips? If they cannot be updated, these chips will be outdated the moment they are made given the breakneck speed at which new and improved models are introduced.