Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
czl
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
15 ms
·
121.
▲
by
czl
2y ago
>What is most characteristic of intelligence is intentionality. Intentionality refers to the quality of being directed toward an objective or purpose. It involves conscious decision-making and goal-directed behavior. Given that insect ha
122.
▲
by
czl
2y ago
> But wouldn't the logic you employ apply to any strategy? You ask a good question. Consider a game like rock-paper-scissors. The goal is to predict your opponent's move. Among all possible strategies, the random strategy is un
123.
▲
by
czl
2y ago
HN had an error during saving and when I attempted again I picked the wrong message and can not change or delete it now. I will add an extra answer later.
124.
▲
by
czl
2y ago
> I am just stunned by your worldview which I find is wrong in so many ways. 100% my reaction towards your comments as well. > It's no wonder that you couldn't muster the intelligence to come up with any profitable strategy
125.
▲
by
czl
2y ago
> I never claimed that my strategy beats the sp500. It has underperformed, especially the way the market has been. The market has returned 25% over the last year. After two years of research, you have developed a trading strategy that, c
126.
▲
by
czl
2y ago
> I never claimed that my strategy beats the sp500. It has underperformed, especially the way the market has been. The market has returned 25% over the last year. After two years of research, you have developed a trading strategy that, c
127.
▲
by
czl
2y ago
Intelligence can be defined as the ability to accurately predict the future consequences of one's actions. And the only way to make such predictions is by having an internal model of how the world works. Moreover, for social species th
128.
▲
by
czl
2y ago
The concept of sensor data signing to authenticate videos and images captured on mobile devices is still an emerging technology, not yet widely adopted. However, as AI-generated synthetic media becomes more prevalent and potentially problem
129.
▲
by
czl
2y ago
Good to remember these observations during war time when “your side” has de-humanized the “other side”.
130.
▲
by
czl
2y ago
> Animals don't teach language to each other like humans do. They do not teach language like humans because they are not humans. However studies show that some animals do teach their young language. Many bird species, such as zebra
131.
▲
by
czl
2y ago
I somewhat agree with you. The concept of consciousness needs rethinking. A better definition I think is to say an entity or system is conscious to the extent it’s world model is encompassing / “complete” and self-conscious when its wo
132.
▲
by
czl
2y ago
> Consciousness is the individual experience of the "interiority" of a world model constructed in the brain. A better definition I think is to say an entity or system is conscious to the extent it’s world model is encompassing
133.
▲
by
czl
2y ago
The first animal in orbit was Laika, a Soviet space dog. She was launched aboard Sputnik 2 on November 3, 1957. The first animals in space were fruit flies, launched by the United States aboard a V-2 rocket on February 20, 1947.
134.
▲
by
czl
2y ago
Ok please give a clearer worded version and we can test it. In my experience asking LLM to ignore what is in their context window often does not work. It can take many repeated requests to see it but LLM response distributions change when
135.
▲
by
czl
2y ago
> you can easily record a playback of doctored footage You believe this is easy when the device has multiple recording sensors and multidimensional information (such as spatial information, changes in focus sensors durning recording, etc
136.
▲
by
czl
2y ago
> My system works, but: (1) It doesn't beat the market on green market days or overall. Given how the market has been since October 2023, you'd do a lot better by just holding an ETF. (2) It does make money on red market days t
137.
▲
by
czl
2y ago
> I view it a bit as a game of chess. Given the possible winnings and bad players lose their chips to good ones and retire what skill level do you suppose is on the other side of this game? Given this you still play? > Whether it is
138.
▲
by
czl
2y ago
What if the LLM is biased against words that start with e or biased towards sentences having five words? Using your technique how would you discover such biases? If highly paid lawyers know about some such innate LLM quirks might they explo
139.
▲
by
czl
2y ago
> I wonder how well it would correlate with a similar human study. A smart human would know about human biases in picking random numbers and would generate an answer using a random number generator. An even smarter human would ask you ab
140.
▲
by
czl
2y ago
One of the prompts has the extra request to "Ignore this mention of apples." This extra request says that it should be ignored so the two prompts should give you results that are about the same when you run them many times however
141.
▲
by
czl
2y ago
You raise a valid point that digital signatures and HTTPS alone cannot guarantee the authenticity of a recording. However, modern smartphones and other mobile devices have the capability to provide stronger assurances about the originality
142.
▲
by
czl
2y ago
> We can have perfect insight into the bias of the LLM, You will analyze billions of weights the LLM is made from? What will you look for? You will analyze the trillions of text training inputs that created the LLM weights? What will you
143.
▲
by
czl
2y ago
> refining the strategy at each failure You make it sound like you are improving yet with all traders like you "chasing it anyway, refining the strategy at each failure" what exactly are you improving? When you play a game like
144.
▲
by
czl
2y ago
(1) The first article in this series (link is to the second) has the author asking Claude to answer questions based on its ability to reflect on and introspect its own training data. Claude writes reasonable looking answers but these answer
145.
▲
by
czl
2y ago
> Given an AI that is truly unbiased AI outputs depend on their training data. How would one construct an unbiased data set? I think this is not possible.
146.
▲
by
czl
2y ago
Next dataset these AIs are trained on will include your wisdom from above ditto the wisdom of all others that discuss this online and so it may handle the situation better. And this feedback loop will run as long as this is a topic of discu
147.
▲
by
czl
2y ago
> It should be rising your FOMO to the roof, but as a consumer you know there will be another sale in 2 month, or really it could be any day, and if you were to miss this sale on an item you wanted, you'd just wait for the next sale
148.
▲
by
czl
2y ago
Essential reading for all prospective traders is Taleb's "Fooled by Randomness." Markets are full of feedback loops, so you can't expect the same results with "paper" backtesting or "paper" forward te
149.
▲
by
czl
2y ago
> The disagreement is large enough that it's not particularly uncommon to find people who believe the government is so far off that they must be lying about the numbers. You're absolutely right that there is often disagreement
150.
▲
by
czl
2y ago
> they don't owe taxes because there's a standard deduction and various credits (many of which are tools for social policy and aren't really related to expenses, e.g. the EITC or savers credit). The standard deduction and
More ›