Acemad (@Acemad_) profile picture

AI PhD with a penchant for software engineering, game development, and graphics design.

Earth
Joined September 2011
Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)Photo by Acemad (@Acemad_)
where i'm at on LLMs: 1. LLMs are NOT going to outmode software engineers, not now, not ever 2. LLMs alone are a dead end for attaining anything like AGI 3. forget about LLMs achieving ASI 4. LLMs will only make good engineers faster, marginally 5. LLMs will remain poor at architecture and design this is 100% because LLMs cannot reason and no, generating a bunch of hidden context is not "reasoning" as always, when the same people are making claims like this, have a look at what they stand to gain and to them, i warn directly: you are going to fall. hard
346
442
66
4,653
308,194
Acemad retweeted
this lives rent-free in my mind. “What if laziness is the habit of thinking about the effort instead of the payoff?”
15
454
8
4,122
131,936
Conor Neill: "Life rewards action. Not intelligence."
18
698
20
4,729
130,026
Anxiety often happens when your level of intelligence isn't aligned with your level of courage. You know what you can do, but you never do it. That's anxiety.
69
747
29
6,394
272,738
Learn to code. Don't let anyone fool you.
1
2
56
Acemad retweeted
Here is a zip file with all the themes: drive.google.com/file/d/15-S… I am not hosting the themes on the site, please disseminate the above link :)
4
6
1,244
Why many teams refuse to do TDD? 1. Don't know how to write structure-shy tests 2. Don't know how to refactor (plus don't see any value in refactoring) 3. What Dijkstra said
9
16
1
75
5,159
I never paid attention to how free (and malloc) works at least in C standard library. malloc takes size and returns a pointer to the start location of memory you requested (slightly more based on the os page size) free takes only a pointer and somehow it can tell exactly how much memory that pointer was allocated and free that much exactly. I asked myself how does free know how much memory it needs to free just from a pointer and I found a nice stackoverflow article. turns out one implementation at least is to store metadata about the allocation in a fixed header right before the pointer. So free technically walks back X bytes to read the header, find out how much memory it needs to free and calls mmap to remove the virtual memory mapping’s essentially for that process. this immediately tells me smaller tiny object allocations can really have massive overhead, this can be even further exacerbated with object orientation where we make everything into an object and allocate/dellocate each object individually. the person answering also mentioned that they implemented an alternative another malloc/free which doesn’t store metadata inline but use fixed size. This makes me appreciate and also question standard libraries and not blindly use them, even if its the bare bone stdlib so much to learn…
13
62
4
600
46,091
The answer to the below question is NO. Artificial intelligence large language models do not seek the truth, nor will they ever understand what truth is. They seek the most popular response to your query, based on training data, the algorithm, and alignment by human workers.
Our “Babbage” podcast puts a listener’s question to @drfeifei, an artificial-intelligence expert: can the problem of large language models’ “hallucinations” ever be fixed? econ.st/40ReXIC 🎧
42
93
7
521
90,104
This is what a great Friday night looked like when I was a teenager
989
13,976
4,637
78,648
9,794,965
This is unacceptable, should not be possible Made possible by @_nonfigurativ_
965
6,444
1,135
55,236
8,562,813
23
195
10
1,429
128,821
Anxiety is when you want something, but don't make the efforts to get it, and are distracted every day by mindless entertainment instead. Peace of mind is when you want something, make a clear plan to get there, and actually steadily get there day after day, year after year.
60
1,203
56
6,782
714,471
If you'd asked me a year ago, superposition would have been by far the reason I was most worried that mechanistic interpretability would hit a dead end. I'm now very optimistic. I'd go as far as saying it's now primarily an engineering problem -- hard, but less fundamental risk.
The fact that most individual neurons are uninterpretable presents a serious roadblock to a mechanistic understanding of language models. We demonstrate a method for decomposing groups of neurons into interpretable features with the potential to move past that roadblock.
54
318
63
2,542
1,951,324
The courage to start. The discipline to focus. The confidence to figure it out. The patience to know progress is not always visible. The persistence to keep going, even on the bad days. That's the formula.
59
1,054
49
4,006
349,546
I don't mind when people are unambitious. It's up to each person to decide that. But it bugs me when the people who aren't ambitious try to bring down those who are.
352
1,291
152
10,276
854,829
Explore the potential of the Object Capture feature with iOS 17. Our second demo video witnesses the transformation of a drinking fountain in Paris into a stunning 3D model, all directly from your iPhone or iPad Pro. Our upcoming AR Code app update promises to further streamline this process, making it simpler than ever to create AR Codes from your scanned objects. Stay tuned for its release at ar-code.com #iOS17 #ARCode #3DScanning #Photogrammetry #AugmentedReality #Paris
182
2,716
586
14,588
1,691,225
"Absence of evidence is not the evidence of absence" -- Dr. Carl Sagan (1934 - 1996)
260
3,006
208
20,849
2,102,898
You wanna get better?
92
33
12
258
25,246
Learning more is a smart person’s favorite way to procrastinate.
165
1,142
168
8,428
779,488