In my experience Claude 3.7 is far superior for coding than Gemini 2.5. I tried it in Cursor and I wanted it to work, as a recent ex-Googler. I repeatedly found it inferior. I think it’s still behind Claud 3.5 for coding.
It would decide arbitrarily not to finish tasks and suggest that I do them. It made simple errors and failed to catch them.
What's the difference between Cline and Roo Code now? Originally Roo was a fork of Cline that added a couple of extra settings. But now it seems like an entirely different app, with it's own website even.
It depends on the task, and prompting feels different.
I've found that sonnet is possibly better at starting things from scratch and frontend code, while Gemini has been able to one-shot difficult problems and fix bugs that sonnet has struggled with.
Switching between them is a frequent occurrence for me.
It might be relevant that I've completely stopped using Cursor in favor of other tools/agents.
I've also created a few "agents" to do specific tasks using Probe[0] as an MCP server, although I'm sure you could create a full-fledged agent with it if you wanted to.
Cursor is likely very tuned for Claude (prompts-wise and all) due to its dominance with 3.5 and now 3.7. Still, Gemini 2.5's tool calling has been pretty poor in my experience which Cursor heavily relies on.
Same here. I've seen some articles and LLM benchmarks that Gemini 2.5 Pro is better than Claude 3.7 on coding, but base on my recent experience of solving code problems with two products, Claude still gave me better answer, Gemini response are more detail and well structured, but less accurate.
Same. I went back from Gemini to Claude yesterday, because Gemini was writing decidedly worse code, at times not even able to stick to Python syntax. Using Aider.
>In the general population overall, Covid is undeniably more deadly than the flu, but only about 3-5x (and I think 3x personally right now).
I don't see how you get close to 3X more deadly than the flu. If 14% of New York state residents have been infected, 20M population, 15000 deaths + another 3000 that are infected now and will die (this disease takes a long time to kill people) you get an IFR of 0.64. If the IFR for flu is 0.05, that makes covid 12X more deadly than the flu.
Lots of people reporting an IFR of 0.5 based on the NY serological data; that is "right censoring" the deaths. It's got to be a bit higher than that. Either way you have covid with 10X the infection fatality rate of the flu. If worst flu years have IFR of 0.1 then covid is still 5X worse than the worse flu seasons. And as contagious as the worst flu seasons too.
So twice as contagious as the flu, 10X deadly. Much more likely to put you in the hospital, and thus overwhelm the hospital system, causing many ancillary deaths. This NYTimes article sums it up: https://www.nytimes.com/2020/04/06/well/live/coronavirus-doc...
At the peak in NY, hospitals stopped seeing heart attacks and strokes, because those people were too afraid to go to the hospital. Many of those people died at home, as supported by the overall death rates in NY.
The narrative put out there by those that look at the recent serology results and say "this proves that this disease was really just a bad flu all along; we can reopen the economy without fear" is just not supported by the data. An IFR of 0.64 and hospitalization rate of double that, like 1.2%, for a disease this contagious, shuts down the economy until we get a vaccine or effective treatment or a Korea-like testing/tracing regime in place.
Yes suicides go up. But this is more than compensated for by other benefits. Overall, people may well be much healthier in a depressed economy. We will certainly see a decline in car accidents.
This is a wrong and dangerous message and it needs to be stopped. Both things are true: a) health workers need masks and b) people should wear masks when they go out.
There are no masks available to the public now. Realistically if you don't already have a mask your only option is to make one, which is not hard. Go to YouTube and find a video. A couple layers of cotton can work and is much better than nothing, at stopping your droplets from spreading if you sneeze or cough.
If you already have a mask wear it when you go out. No health care worker wants your mask. Wear goggles too.
And yes masks help prevent respiratory infection which is why health care workers wear them.
Stop with the "you shouldn't be wearing one" nonsense. It is false and dangerous.
I live in New York City. I haven't left my building in weeks, but I can see people on the sidewalk outside of my window, and I see other residents of my building in the lobby when I go (briefly!) to pick up laundry and mail.
Many of them are wearing surgical masks. Not N95 masks, true, but not home-made masks either. I don't know where they're getting the masks, but clearly they have some.
Every time I see this, I honestly kind of want to scream at them.
If you want to wrap a piece of fabric around your nose and mouth, by all means, it certainly can't hurt. But if you have a supply of real masks somewhere, you should donate them.
I keep a box of surgical masks, since I'm one of the weirdos who wears one when sick even in the US! I've got 3. The box is years old, they're expired, they're not sterile any more, and I'll keep re-using them until they fail (the elastic stretches). As desperate as hospitals are these still aren't donatable.
To find the units in a drink you multiply the ABV by the serving size in litres.
One 200 ml glass of wine at 13% ABV would be 2.6 units.
One 330ml bottle of beer at 5% ABV is 1.65 units.
Most people struggle to estimate how many units of alcohol they drink. The "one unit == one drink" thing comes from when units were introduced, when a glass of wine would be 125 ml and it would only be 8%. Those days are long gone.
Maybe we should go back to those days. When I drink wine at home, I haven't measured it, but I'm quite sure I don't drink more than 125 mL at a time as I can easily get more than 6 glasses out of a standard 750mL bottle.
That's what a lot of sources in the US say, but, when they say, "drink", they mean "standard drink", which is this quasi-mythological beverage that one never encounters in real life.
I honestly think that it's negligent on the part of public health folks to continue the practice. Saying "Drinking more than 2 drinks in X time will cause you to be legally impaired" is actively encouraging people to unwittingly drive drunk in a culture when a typical ABV for beer is about twice what it was when those standard were originally calculated, and non-highball cocktails typically contain at least two shots of liquor.
http://www.tynker.com. My daughter started with Scratch Jr. When she turned 9, we got her a chromebook for her birthday, and I read about this site in a WSJ article. She took to it right away. Kids can progress through extensive programming-puzzle tutorials and make their own projects, using a "block" language, python, or Javascript. They can switch between the three supported languages within projects. She is still most comfortable with the block style, given that it abstracts away free-form syntax, but has started working through the python tutorials. A year later she has made dozens of her own projects (mainly simple games) of increasing complexity, which can be published and shared with other kids. She now understands relatively complex concepts like generating random behavior, method calls, variables and even physics engines. Highly recommended.
Revealing quote from Baez in this article: "I’m ... eager to dig deeper into open reaction networks. They’re a small but nontrivial step toward my dream of a mathematics of living systems. My working hypothesis is that living systems seem ‘messy’ to physicists because they operate at a higher level of abstraction. That’s what I’m trying to explore."
Worth reading for this sentence (p.5): "The core obstacle to an integration of gravity in the context of quantum field theory is the occurrence of untreatable infinities in calculations of particle interactions due to the possibility of point particles coming arbitrarily close to each other."
First time I've heard the problem of integrating gravity with QFT explained that way.
"untreatable infinities in calculations of particle interactions due to the possibility of point particles coming arbitrarily close to each other."
This is actually very misleading. As mentioned in other comments, QED has "infinities in calculations of particle interactions due to the possibility of point particles coming arbitrarily close to each other," but these are treated via renormalization. QED (Quantum ElectroDynamics) is the QFT (Quantum Field Theory) based on Maxwells equations of Electromagnetism. Maxwells equations are linear partial differential equations and when converted to a quantized form, via QFT, the infinities can be dealt with by Renormalization. Einstein's General Relativity equations are non-linear partial differential equations and when converted to a quantized form, via QFT, the infinities cannot be dealt with by Renormalization.
The quote makes it sound like classical point particles "coming arbitrarily close" is the problem with quantizing gravity(General Relativity), but the same problem exists in QED and yet QED yields a viable theory through Renormalization. The real difference is that Maxwell's equations are linear and Einstein's equations are non-linear. OTOH, the non-linearities in Einstein's equations do not become important until the particles are very close.
I remember hearing about this issue—or perhaps a related one in another formulation/area of quantum theory—where the 'infinities' had to be treated by something called renormalization[0].
Every time I hear about these problematic 'infinities,' I can't help but think of a novice programmer looking at the console output of their failing program, "it gave me all these weird symbols and says something about an 'exception'".
Maybe what's confusing to me is that 'getting infinities' in this way is somehow normal and not indicative of a bug, maybe? Otherwise, how can they be brought up in this way without the conclusion being, "seems like we got it wrong, time to try something else."
The problem is that the gravitational infinities one gets in QFT are non-reonormalizable. Renormalizable theories must follow a very strict criteria (and even then they are rather difficult to grasp if you don't really put a lot of thought into it).
I assumed that was the case, but it doesn't alter the impression I get about the situation. That still sounds to me like, "we can only apply this fix when instances of the problem follow certain strict criteria".
IIRC Feynman was also concerned about the infinities as well, but ultimately pragmatic. The infinities bother me as well. But, on reflection classical point particles produce forces that approach infinity as they become close, so the quantatization reflecting this seems reasonable.
Please see my other comments.
Thanks for pointing this out. I may have overlooked this article otherwise. As a non-physicist I can't say I've ever come close to understanding the problem of integrating gravity with QFT and this statement alone really helped.
If you follow your Wikipedia link and then follow the link for Green's function two times, you come to the article that points out that Green's function methods only work on linear differential equations. The multipole method you refer to is based on the Green's function and only works for linear differential equations. Einstein's equations are non-linear. Trying to quantize a non-linear theory leads to problems that no one has yet resolved. You should also note that non-linear differntial equations do not support superposition; that is, the solutions cannot be added to yield another valid solution. They are not additive. So, the method you suggest fails because it does not meet the criteria you refer to.
Gravitational interactions become stronger with decreasing distance.
Take the Newtonian gravity model, if force is proportional to 1/distance^2, as distance approaches zero, the force approaches infinity.
You can simplify all points in a region as being one point for points that are far away, but that doesn't help with the infinite force problem with close points.
>The web is one of the great stories of backwards compatibility
Are you serious? Flash is being dropped eventually, npapi plugins (especially wrt Java) are out, the blink tag is gone, Firefox extensions are being dropped for a worse system, breaking changes to JavaScript, the list goes on.
The attitude you express towards Misophonia is one of the reasons why having this disorder is such a burden. It has nothing to do with table manners. There is nothing rational, or for that matter culturally relevant, about the emotional reaction I have to someone eating an apple next to me. The visceral, angry emotional response I feel to that sound has nothing to do with culture.
That sound is not the only one that triggers my response, but I chose it because of this argument: I don't think anyone would argue that there is something culturally wrong (at least in western culture) with eating apples around other people. And there isn't really a way to eat a whole apple without making a sound that makes me irrationally furious.
I can't stand others eating an apple, but I can eat one without and issues. The same behavior holds true for playing with loose coins or cellophane wrappers.
That's right - I don't feel anything when I am eating myself, or even when I am at a lunch table with others munching with me. It's only when I am not the one eating. I also get irritated when I am on the phone and the person at the other end is chewing or lapping.
Also, I don't think it's cultural and I suspect it might be hereditary.
Interesting. What if you listen to a recording of yourself eating? While you're at if see how much you can distort it before it stops or starts to bother you. E.g. Played backwards, at double speed, half speed, etc.
Same here. I think this is evidence of genetic factors - i.e. my dad has a mild case and it manifested in him being very emphatic about training us not to make sounds while eating.
It would decide arbitrarily not to finish tasks and suggest that I do them. It made simple errors and failed to catch them.