vessenes

9 ms·

151.
by vessenes 4mo ago
Can you talk more about RISC Zero? Does it require a TEE of some sort? I had trouble finding a quality mid-detail spec of how it works; lots of marketing materials basically.
152.
by vessenes 4mo ago
I believe there's some data that they will have different responses if the names signify different cultural / race / gender affiliations. Here be dragons.
153.
by vessenes 4mo ago
But those are VERY different types of assistant. It is correct behavior that you would get different outputs in this case.
154.
by vessenes 4mo ago
Hi there. I've done a bit of work on specifying human-centric identity goals for the internet over the last 10 years. May I suggest you look at Microsoft Vega? https://www.microsoft.com/en-us/research/blog&#x2
155.
by vessenes 4mo ago
Agreed on the desire for continuous behavior. That said, in a modern LLM, is this hash analogy accurate? I would be surprised if a single letter changed most zero temp force ranked outputs. E.g: “Where is the Eiffel Tower Located? One word
156.
by vessenes 4mo ago
Don’t underestimate the impact of AOL - the main experience boomers and greatest generation had of the early internet was moderated and mediated. My parents were not on IRC.
157.
by vessenes 4mo ago
To be clear, temperature 0 is deterministic and will produce the same output for exact duplicate inputs, across all seed choices. Provided: * If it’s MoE we are talking about, that the duplicate inputs are for the whole batch (yes, your b
158.
by vessenes 4mo ago
I believe cerebras is one wafer, not deeply stacked, each core is like half memory half compute by area.
159.
by vessenes 4mo ago
Minkowsky, cool design! Question - the ASIC designers I've worked with over the years have been fairly adamant that integrating memory on package interspersed with logic is very difficult; the general statements run like "those de
160.
by vessenes 4mo ago
Missing some recent models on that list, but I think most crucially, the harness is fixed —- one of the major learnings of the last few months is that harness and eval (“looping” and support / tooling around it) is really critical. I
161.
by vessenes 4mo ago
Very cool. I’m reminded of Wolfram’s pitch that neural nets are a search through the very broad computational complexity of the function space they describe; he did a little work to show that you could find similar behavior in other functio
162.
by vessenes 4mo ago
Yes, I had the same question. I don’t think so, as currently designed. It trains to specific points / classes in an embedding space. They didn’t discuss how one might go to non-trained points in the paper as far as I could read, and th
163.
by vessenes 4mo ago
Also true: all computing is analog computing.
164.
by vessenes 4mo ago
The article says base M7 memory bandwidth is targeted at 240GB/s. M1 had 70 GB/s, M1 Pro: 200, M1 Max 400, M1 Ultra 800. Modern RTX 6000: ~1,600 or so. If we get a 1,200-1,500 GB/s bandwidth M7 variant in late 2027 with 512GB
165.
by vessenes 4mo ago
My understanding is they are largely an IP business. That said this release mentioned an ASML machine on prem, so?
166.
by vessenes 4mo ago
Aren’t you supposed to be on vacation?? Don’t waste it all on HN! :) Hope your summer is going well.
167.
by vessenes 4mo ago
Really nice result, congrats to the Aisle team. What stood out the most to me here was their pitch that harness currently matters most, over and above a specific model capacity. That’s one of my conclusions reading cloudflare’s Mythos debri
168.
by vessenes 4mo ago
I don’t know the methodology at this site, but I noted with interest they rate grok build slightly below Meta’s offering(!) I would anticipate we will see aggressive improvements / new models at least as the Cursor integration comes to
169.
by vessenes 4mo ago
SpaceX is cashflow positive. Buying compute that others pay a premium for is not over-buying. Note that $85b GOOG raise is for … building compute they don’t have. The SX bond issuance reportedly has $90b in demand; they took $25b. With all
170.
by vessenes 4mo ago
I’m sure from our takes on some of these we have pretty different ideological bents; hello my friend with different views! :) To get down my take over virtual beers with you, The CBPP leads that page you linked with “The years from the end
171.
by vessenes 4mo ago
I’d like to pick on a couple points here that I think are not just nits: > more concessions to the working class. The 1950s were incredibly prosperous as a result This is absolutely not a common read of the boom in the 1950s. The common
172.
by vessenes 4mo ago
It's really not the same thing. Read the cloudflare blog about using Mythos. Mythos is important and notable because of the harness and self-direction. It's not necessarily a way stronger bug finder, but it was trained to do the e
173.
by vessenes 4mo ago
Actually, ELO rankings done blinded on models do vary: https://the-frontier.app , that said, your point looks accurate as far as 5.3 - 5.5 on this chart, 40 to 50 point ELO gain. I find I have to argue with 5.5 less than 5.3, and
174.
by vessenes 4mo ago
I think you'll find other labs are racing to get you something while Anthropic works through their issues. So, yes, give it a few months, you'll have something equivalent from someone somewhere in the world.
175.
by vessenes 4mo ago
I thought about it, but don't think I'll push the button. I have a falcon nw gaming rig in my living room right now running windows / steam big picture and an NVIDIA 3060Ti -- and it's .. fine, but long in the tooth. I w
176.
by vessenes 4mo ago
Wow, there is a lot of negative gatekeeping on learning Japanese in this thread. I recommend people chill out. Dan, this is cool, and I can tell you enjoyed writing this up and thinking it through as you wrote. Don’t let the haters get you
177.
by vessenes 4mo ago
Much less than the 2 shares of apple stock you could have bought :) I’d be upset too.
178.
by vessenes 4mo ago
Boy that is some ugly C89. Like truly terrible. Lots of ways to ‘port’ something like this, but I think if you miss the instructional / simplicity angle, you have not created a good translation.
179.
by vessenes 4mo ago
sadly this is so mangled by the clankers that wrote it, it's not helpful. Neal Stephenson's Anathem imagines a punishment in his dream monastery as a book you have to memorize with subtly incorrect information -- I'm sure rea
180.
by vessenes 4mo ago
Tell me it’s a bad dream?

More ›