alain94040

9 ms·

91.
by alain94040 3y ago
Full array in L1 is not a typical scenario for binary search. Binary search is usually for large data sets, that are in DRAM.
92.
by alain94040 3y ago
I don't get it. The problem with binary search and branches is not the branches themselves, it's the fact that until you have done the comparison, you don't know which memory location in the array to fetch next. It doesn&#x27
93.
by alain94040 3y ago
Also, the read ports are asynchronous, which is bad: assign rv1 = registers[rs1]; assign rv2 = registers[rs2];
94.
by alain94040 3y ago
It goes beyond being inefficient. It's the wrong mental model. Chip designers intuitively visualize interconnected blocks of logic. They draw rectangles with wires between them. To learn this intuition, writing spaghetti Verilog code t
95.
by alain94040 3y ago
Correct. The code shown is only good enough for RTL simulation. It won't help actually generate a chip with multiple read and write ports to a register file.
96.
by alain94040 3y ago
I’ve spent nearly twenty years blogging, giving talks and releasing open source code. [...] I’ve interviewed hundreds of people [...] The vast majority of candidates have little to no evidence of creativity in public at all Are we sure th
97.
by alain94040 3y ago
Correct. Here’s a pro-tip that also applies at work, when you email your manager: make it as easy as possible for the person to help you. If you ask for an intro to someone, write an email that is ready to be forwarded, as is, and already h
98.
by alain94040 3y ago
The key point: A key architectural feature to achieve this is the ability to fit all model parameters inside the on-chip SRAMs of the chiplets to eliminate bandwidth limitations. Doing so is non-trivial as the amount of memory required is
99.
by alain94040 3y ago
I found the paper didn't live up to its claims. It said: the ever-changing behavior of these SoCs is also visible via internal measurement sensors, allowing us to distinguish between executed instructions, and even different operands
100.
by alain94040 3y ago
I'm quite disappointed in Knuth's dismissal. As the "king of algorithms", he of all people, should appreciate the leap. To put it another way, imagine if one of his students came to him with this question: I want to writ
101.
by alain94040 3y ago
This thread may be as good as any to ask: the code I write has to deal with a huge amount of state. To summarize, my code receives millions of "change commands", which update state in memory. Data structures are hash tables, lists
102.
by alain94040 3y ago
There are 3 places where people can charge their car: at home, at work, and at the store. If you can't charge at home because you park on the street, that still leaves two opportunities to charge. Another way to think about it is where
103.
by alain94040 4y ago
I don't see ads as a big problem with ChatGPT. You could put a side-bar on the right and on the fly recommend products relevant to the on-going conversation. The cost of computing these ads would be a lot more than today's keyword
104.
by alain94040 4y ago
Yes. I think of 5 people I contacted over 3 years, one got hired. (and it was a very successful hire)
105.
by alain94040 4y ago
This long article is a great example of why I believe GPT has a strong future. The article spends several pages to explain hexadecimal, bases, etc. Probably some fraction of the audience already knows that. In that case, the article should
106.
by alain94040 4y ago
Another interesting case is possessive in French: une hache - ma hache une horloge - mon horloge
107.
by alain94040 4y ago
Those new language models are "Google killers" because they reset all the assumptions that people have made about search for several decades. Imagine that people start using those chat bots massively as a replacement for Google se
108.
by alain94040 4y ago
Some back of the envelope math: - there is a branch instruction every 10 instructions or so - if you can't predict, assume you guess wrong half of the time So every 20 instructions or so, you take the wrong branch. Sounds like it could
109.
Benefit of doing Slack “interviews” is that you get a transcript of the session(infosec.exchange)
1 points by alain94040 4y ago|0 comments
110.
by alain94040 4y ago
Sure. But if you combine the understanding that this chatbot has with a Wolfram Alpha backend, you could build an even more amazing system. I'm sure someone is working on hooking up language models to math backends (anywhere from a sim
111.
by alain94040 4y ago
I was quite impressed by its capability to correct itself. My test went like this: Q: Do not use the imperial system Q: [some question which involves distances] A: blah blah 1500 kilometers, which is 1000 miles blah blah Q: I told you not t
112.
by alain94040 4y ago
It may help a 10X engineer. The code is not what's hard, it's figuring the correct data structures that matters (also known as architecture). Once you got the data structures correct, the code mostly writes itself, and I can see a
113.
by alain94040 4y ago
Where are you located?
114.
by alain94040 4y ago
I'll chime in to say that I was very happy with @grellas's firm in Silicon Valley. I found them the right mix of startup-friendly and competent in tech issues. (grellas.com).
115.
by alain94040 4y ago
Yes: GB vs Gbit. So an order of magnitude.
116.
by alain94040 4y ago
Are you sure? A quick Google shows: Wifi 6 bandwidth: 9.6 Gbit/s M1 SSD read speed: 6000 MB/s So the read speed is about 6X faster than the Wifi 6 bandwidth.
117.
by alain94040 4y ago
Assuming you are correct, it still means that any startup that has an idea for an innovation in the phone cable space will first need to worry about how to convince the European Commission to approve it. That's quite a road block for a
118.
by alain94040 4y ago
The primer is simple: if you notice that your code is using too many if statements, especially nested ones, and you are starting to wonder if you are covering all alternatives and cases, then using the rigor of a state machine may help. Y
119.
by alain94040 4y ago
Thanks to Moore's law, you can assume that DRAM capacity will double every 1-3 years. Every time it doubles, you need one more bit. So if we use 48 bits today, we have 16 bits left to grow, which gives us at least 16 years of margin, a
120.
by alain94040 4y ago
Cost of sales includes marketing expenses. Possibly also operations cost.

More ›