jordz

I think a sub-set of people see this coming, I also think it isn’t just fine tuning open weight LLM models. A few people I know who are thinking along the same lines with architectures like BERT etc.

That being said it’s much easier at the moment to continue to use the frontier providers for most general tasks, that is the argument I’ve heard.

For creating these types of fine tuned local models, on constrained hardware for inference, I do think this is the way to go for specific tasks too!

Opus 4.6 was the sweet spot for me as a thinking partner specifically.

I use these models for coding, but also a lot of product, commercial, financial and architectural work where I’m trying to develop something half-formed. 4.6 was unusually good at understanding what I was trying to get at, playing it back cleanly, getting the nuance, and extending it without bastardising it as the conversation was drawn out.

It could make useful connections without constantly trying to manufacture an insight.

5.6 Sol is genuinely excellent at the creative part, and in some cases better than 4.6. My issue is convergence to get to a point, a final point. As you try to distil an idea, it often invents new terminology for concepts you’ve already established but its so subtle you have to really keep track of it. The vocabulary and idea tree keep expanding when what you actually want is to collapse everything down to the few things that matter.

Opus 5 has the same problem for me that barrkel said, the prose is often so elliptical and I just want it to tell me it and get to the point than making me dance around what its trying to tell me.

I don’t think 4.6 was necessarily the most capable model (compared to Fable) for long horizon task delivery, and Opus 5 is much more Fable like, it's fiercely determined to get through the task list .

4.6 just felt unusually well calibrated to my way of collaborating and its ability to understand, extend and then compress my thinking without constantly imposing some random walk.

Can someone please explain or link to some information about how models are merged? Is this genuinely merging weights mathematically or some kind of distillation (presumably not if they’ve done zero training as the post suggests).

I work in the industry making hardware and software for large scale commercial and grid scale storage.

There are several challenges with this, safety, thermal runaway, and life cycle of the asset which has a limited amount of cycles.

Also the architecture of the system for the AC inverters and the DC side can come from very different places in the supply chain and aren’t as vertically integrated leaving you in a position where you can’t actually make this work without compromising something in the supply chain. That being said we are talking about a LOT of energy in these systems and to dissipate that much heat you’d need a load bank.

It was edited. It definitely had -274 (:

Does anyone else find it strange that craters on the moon can reach a temperature lower than absolute zero, according this article anyway? =)

Cool. We have worked with a customer on this exact thing and deployed an edge and cloud controller that orchestrates control changes based on the all the turbines. Such a great project!

Lost my 635 day streak last week, offered £5.99 or something to repair it. I was pretty gutted but actually on reflection, the gamification just added undue stress. Duolingo is great for vocab and repetition, it was only with a tutor that I really became conversational.

Also agree. However, I do yearn to relive the days where one can load up a web page and start reading content, without being jarred by pop-ups and ads that inadvertently move the content post load (which of course happens 3 seconds later..).

Ahh this has made my day. I haven't seen a number of these, they all look exactly like the kind of film I'd love to try.

Thank you for sharing!

"Amazon does not charge any fees to join Amazon Sidewalk"

Not to the end customer anyway. To organizations and developers building on top of the APIs...

Completely against this idea though its a shame its not... "Get paid for hosting a network using the devices you purchase from us"*

Lastly, it is a cost albeit negligible to the end user who installs the device (power, bandwidth, etc).

*I know it wouldn't amount to anything but its a interesting way to think about it.

We too are also doing something similar. We've just started the move to timescale for real-time energy and sensor data from industrial assets. We have a single customer with about 10TB of data and growing from 2 years worth of real-time monitoring which is stored in a mixture of table storage and SQL. Timescale on PG seems like blessing for our future plans :)

There is definitely a need for better technology for start-ups in this space.

Having said this, I don't know what this product does from the website.

We pay ~$80 a month for 2 users to a financial planning application that integrates with our accounts (Xero) and it works well. Our revenue is 7 digits annually.

As a start-up founder, I appreciate what you're trying to do with your model but I kind of like, you know, just knowing the fixed costs regardless of our revenue.