ttul

27 ms·

421.
by ttul 2y ago
Yes, for sure intertial and star guidance has been in use in ICBMs forever. But if the inertial guidance is far more accurate, perhaps this changes what you can do with conventional ICBM munitions. For instance being able to target somethin
422.
by ttul 2y ago
If the R&D is going toward intellectual property that is owned by the company, yes. And it doesn’t matter if the company is foreign or locally owned. The program is designed to foster companies hiring engineers and scientists to work in
423.
by ttul 2y ago
If Sandia is revealing this, they probably have a secret version that is already being loaded into cruise missiles and ICBMs. Seeing what’s happening over in Ukraine, it has to be clear to every military in the world that GPS is no longer o
424.
by ttul 2y ago
I suspect that his figure on the number of hours is somewhat cooked up and biased. Did he cite a reliable and reasonable source of data that we can all consult to check the veracity of this claim?
425.
by ttul 2y ago
I kind of want discovery to happen in this situation.
426.
by ttul 2y ago
This is one reason why some companies have located engineers in Canada under subsidiaries. Canada not only allows you to deduct R&D costs as an expense, but there is an extremely generous R&D tax credit that yields a negative tax ra
427.
by ttul 2y ago
To whit: OpenAI was until quite recently investigating having TSMC build a dedicated semiconductor fab to produce OpenAI chips [1]: (Translated from Chinese) > According to industry insiders, OpenAI originally actively negotiated with TS
428.
by ttul 2y ago
While it's true that language models are fundamentally based on statistical patterns in language, characterizing them as mere "probabilistic syllable generators" significantly understates their capabilities and functional int
429.
by ttul 2y ago
For people who didn’t grow up in the 1980s and 1990s, these will be key insights!
430.
by ttul 2y ago
Thanks for the critique. Here is how I would answer their question myself: o1 is far more than just CoT mechanics. It relies on a specialized model or collection of models that offer new capabilities to make CoT work far better than it work
431.
by ttul 2y ago
Pasting from my Perplexity page on the topic: The core innovation [1] of o1 lies in its ability to generate and refine internal chains of thought before producing a final output [2]. Unlike traditional LLMs that primarily focus on next-toke
432.
by ttul 2y ago
I have to agree with you here. OpenAI may be playing for competitive advantage more than for the good of humanity by hiding the results.
433.
by ttul 2y ago
OpenAI’s answer to this would be, “Okay then, don’t use it.”
434.
by ttul 2y ago
It seems OpenAI has decided to keep the CoT results a secret. If they were to allow the model to call out to tools to help fill in the CoT steps, then this might reveal what the model is thinking - something they do not want the outside wor
435.
by ttul 2y ago
That “main engineering effort” will go on forever. People neglect to note that everything is constantly changing. Just like the roof on your house, if you don’t upgrade your components regularly, eventually you will face a huge rewrite when
436.
by ttul 2y ago
I've given this a test run on some email threads, asking the model to extract the positions and requirements of each person in a lengthy and convoluted discussion. It absolutely nailed the result, far exceeding what Claude 3.5 Sonnet w
437.
by ttul 2y ago
I mean, $7,000 a month isn’t nothing. But it’s not a lot. Certainly not enough to justify a seven month engineering effort plus infinite ongoing maintenance.
438.
by ttul 2y ago
Indeed. Nobody would describe a 150 billion dimensional system to be “simple”.
439.
by ttul 2y ago
I believe that the system prompt offers a way to fix up alignment issues that could not be resolved during training. The model could train forever, but at some point, they have to release it.
440.
by ttul 2y ago
If your startup has a chance at an exit in the >$5M range, then you should definitely have a personal accountant analyze your situation. Capital gains taxes start to crush your dreams when you are into that realm, and structuring things
441.
by ttul 2y ago
I suggest there is value in optimizing your own personal tax structure, though. In many countries, setting up a family trust to own your company shares provides enormous upside if you have a large exit, because your family members - includi
442.
by ttul 2y ago
Ahhh so that’s why my system started sucking in May. I actually started looking at replacing everything. I hope their CEO is losing sleep.
443.
by ttul 2y ago
I'd respectfully disagree with this characterization of LLMs. While they certainly excel at pattern recognition, calling them mere "autocomplete machines" vastly undersells their capabilities. LLMs demonstrate complex reasoni
444.
by ttul 2y ago
More seriously, though, budgeting models benefit from the ability to express a random variable at each input so that you can graph the possible range of outcomes.
445.
by ttul 2y ago
You need a random number generator called “Having Kids” and another one labeled “Divorce”. This would complete the simulation. /s
446.
by ttul 2y ago
I’ve been favoring Tailscale lately for establishing magical access to machines at home. Because it permits two-factor authentication based on Google and other systems, it seems more secure than just having things exposed via public IP. Tha
447.
by ttul 2y ago
Working as an intern in 1999, I was tasked with writing a standards document for VoIP telephones for the TIA working group working on this space. The standards organization had a Word template that everyone used, but as an engineering stude
448.
by ttul 2y ago
… and that situation will persist until other vendors release consumer GPUs with significant VRAM. Nvidia craftily hamstrings the top consumer GPUs by restricting VRAM to 24GB. To get a bit more costs 3-5x. Only competition will fix this.
449.
by ttul 2y ago
Love?
450.
by ttul 2y ago
I didn’t want to read this, so I asked Claude to ELI5 it to an EE-graduate level. I hope this summary is useful: CURLoRA is a new way to fine-tune large language models (LLMs) that aims to solve two main problems: 1. Catastrophic forgetting

More ›