The difference here reminds me of the distinction between the regular Paxos algorithm and the Byzantine Paxos algorithm.
19,975 karma · joined April 8, 2014
I love Big Tech more than I love startups. Feel free to disagree with me on that.
The difference here reminds me of the distinction between the regular Paxos algorithm and the Byzantine Paxos algorithm.
A lot of specs contain these optional limitations. The C standard allowed implementations to truncate variable names to 31 characters if they have external linkage and 63 if internal linkage. Would you be happy if your implementation actually does that? Of course not. Neither LLVM nor GCC does that. That’s because people writing compilers also hold themselves to a higher standard. Now imagine this kind of attitude being pervasive.
The most grating thing is requiring every workload to be in a dedicated VM. You are throwing so much performance away. What kind of threat model would justify this? Usually there’s no threat model analysis at all.
Banning opensshd is again mostly a fool’s errand. It’s much more productive to design a system that writes the authorized_keys file read by opensshd. For example you can design a system that requires peer approval before your own key gets written to authorized_keys.
Besides Solarized Dark, GitHub’s Soft Dark theme is quite well done, but unfortunately it’s not the default for a logged-out user with dark mode. So with dark mode I need to log in to GitHub to get acceptable colors.
Now that I think about it, I need a feature to limit the darkness of colors anywhere on the page, background or text except images. Remap #000 to #222 and #fff to #ddd.
> It’s precisely because Adaptive Dark Mode is happening deep within the WebKit browser engine that we are able to find creative ways to solve the most complex challenges, apply the transformations with great speed and power efficiency
Don’t you love it when browsers lie to websites to enhance the user’s privacy.
So yeah the blog post says Google Search itself will be entirely off until being turned on. There’s nothing in this blog post to indicate whether Google Search can be turned on without also turning on Gemini. Although I hate the NYT, the NYT is quoting someone from Google saying there won’t be separate controls for Gemini and Google Search.
I should note that the original authors did not allow Astra to write any significant code to construct the weights. I think this is unfair. If you allow strided assignments to numpy arrays, you should probably also allow the use of functions like np.linspace, np.repeat, and np.tile.
Oh and Astra wrote this (https://github.com/RikiyaT/lm_understand-logs/blob/4d09ba6d6...):
W1[63:120:6, 5] = 10
W1[123:180:6, 5] = 20
W1[183:240:6, 5] = 30
W1[243:300:6, 5] = 40
W1[303:360:6, 5] = 50
W1[363:420:6, 5] = 60
W1[423:480:6, 5] = 70
W1[483:540:6, 5] = 70
W1[543:600:6, 5] = 90
I think the second 70 there is probably a typo; it looks like it should be 80 instead.Maybe you are thinking of high school or middle school textbooks?
Of course time will tell whether this brings failure rate down to be negligible. But I feel pretty confident.
Furthermore, having good textbooks papers over the fact that in the U.S. the qualification and ability of teachers vary so much across poorer and richer school districts. A smart student in a poor school district can use the textbooks to a greater advantage.
Enterprises care far more about vendor support than the typical Linux hacker at home.
Yup. This is what Terence Tao has been saying about digestion of results for several months now. Mathematics has become a field where results themselves matter less, the social aspect of attracting others to understand your results and explore new problems has become more important.
In a sense it is a little bit like software engineering, where writing code can sometimes be delegated to AI, and is arguably less important than understanding what the customer really wants from this software.
For Christ’s sake a natural disaster just struck and you want to live like nothing happened? The entitlement is unbelievable.
I find the risk-award very clear: if you are not deficient, taking normal* amount of Vitamin D supplement has no adverse effects; but if you are deficient, the risk is quite high.
*: The normal amount is 600 IU. The recommended upper limit is 4000 IU. The gap is wide.
> Alert noise reduction and intelligent grouping, runbook execution for known issues, log pattern detection and anomaly flagging, and basic root-cause suggestions from historical incident data, and auto-remediation for well-understood failure modes like restarting a crashed pod or scaling up a service that’s running hot, are all fairly automated today.
Let’s stop here and ask, why do we not fix the underlying issue? Alert noise? Design better alerts. Runbook execution? Make it a cronjob. Known issues? Why aren’t people fixing these known issues? Root cause suggestion from historical data? Sorry if root cause A is in the historical data you expect problem A is fully solved so it wouldn’t appear again in the next incident. Restarting things? Why doesn’t your health monitor restart things automatically?
SREs also need to climb up the abstraction ladder. Instead of using AI to assist with grunt work, use AI or human thinking to build more reliable systems with fewer incidents. Every incident should be novel.