The logs from Sweep(which contain snippets of code) are logged for debugging purposes. We don't train on any of your code. These will only be stored for 30 days. We send this data to OpenAI to generate code.
We're using the OpenAI api, and OpenAI has an agreement stating they will not train on this data and will persist it for 30 days to monitor trust and safety.
We index your codebase for search, but we use a system that only reads your repo at runtime in Modal. This runs as a serverless function which is torn down after your request completes. Here's a blog we wrote about it! https://docs.sweep.dev/blogs/search-infra
We need a way to store the computed embeddings, because the function doesn't persist any state by default, so we use Redis. But we don't want to store the actual code as the key, so we hash the code + add some versioning. Because it's a cache, it supports concurrent writes + reads, which a lot of vector dbs do poorly.
So the actual code is only accessed at runtime (using the GitHub app authentication to clone the repo), and we also build the vector db in memory at runtime. It's slow(redis call, embedding the misses, constructing the index), but 1-2s is negligible in the context of Sweep because a single openai call could be 7s+.
And one nice feature is that when you have Sweep running on 10+ branches (which probably share 95%+ of the code) we just use the cache hits/misses to automatically handle diffs in the vector db. It's super easy to setup, we don't need to manage different indices (imagine a new index per branch), and it's very cost efficient.
It's also a factual error. There have been longevity genes discovered. Cynthia Kenyon famously doubled the lifespan of a C. Elegans by mutating daf-2: https://www.nature.com/articles/366461a0
I thought it'd be too much in the weeds. The UTexas paper goes into their prescribing data and comes out with the conclusion that calcineurin inhibitors are the only common factor in their patients who don't get dementia.
Rapamycin also does not have any evidence of efficacy on neurodegeneration, while there's evidence for CNI on other forms of neurodegeneration.
Cyclosporine has some very serious long-term toxicity (especially renal dysfunction and hypertension) and the availability of newer biologic agents has restricted the use of cyclosporine to patients who have not responded to conventional treatment.
*Mild tremor is common with both cyclosporine and tacrolimus use, occurring in 35 to 55 percent of patients [ 79,80 ].
*Rarely, patients develop severe headache, visual abnormalities, and seizures. This syndrome is associated with acute hypertension and resembles hypertensive encephalopathy [ 77 ].
*Posterior leukoencephalopathy is usually seen on brain imaging [ 82 ].
*Cyclosporine may contribute to bone loss after organ transplantation [ 88 ]; this effect may be due to the induction of high bone turnover.
*Cyclosporine can cause hyperkalemia and hypomagnesemia, via effects on renal tubular function. Hyperuricemia and exacerbation of gout are well-described with cyclosporine.
[77]. Schwartz RB, Bravo SM, Klufas RA, et al. Cyclosporine neurotoxicity and its relationship to hypertensive encephalopathy: CT and MR findings in 16 cases. AJR Am J Roentgenol 1995; 165:627.
[78]. Eidelman BH, Abu-Elmagd K, Wilson J, et al. Neurologic complications of FK 506. Transplant Proc 1991; 23:3175.
[79]. Randomised trial comparing tacrolimus (FK506) and cyclosporin in prevention of liver allograft rejection. European FK506 Multicentre Liver Study Group. Lancet 1994; 344:423.
[80]. A comparison of tacrolimus (FK 506) and cyclosporine for immunosuppression in liver transplantation. The U.S. Multicenter FK506 Liver Study Group. N Engl J Med 1994; 331:1110.
[81]. Wijdicks EF, Wiesner RH, Krom RA. Neurotoxicity in liver transplant recipients with cyclosporine immunosuppression. Neurology 1995; 45:1962.
- They focus on avoiding blood vessels. It is up to the software to learn the mapping, after implantation.
- They talked about flexibility, slack of electrodes, and pig head-butting. :)
It is used as a type system for code generation and high performance calculation with discrete differential goemetry sub algebras. More information is at https://grassmann.crucialflow.com or also at https://bivector.net
It's up to the registrar to make the rules. If nic.io is fine with people outside of that territory to use their tld then it's okay.
Some TLDs do indeed have stronger requirements. I remember that for a long time getting a .fr was a bit of a pain because you had to prove that you were a french citizen or a french company IIRC.
Agree. Most of program languages are context free. Human language is mostly context dependent. The auto completion and auto suggestion are the tools to close the gaps of the user experience.
Moreover, human communication is continuous and conversational. Programming is not. Most of existing code editors are not designed to have a conversation. Jupyter notebook is close but not there yet. I bet with a conversational agent style code editor plus a good auto suggestion and auto completion feature, we don't need to invent technology to use natural language to communicate with a machine. We just need a well designed formal language with precise and concise syntax.
Some people write news conforming to Google translation's performance just to make sure the story can be auto translated to many other languages. Most of these stories have pretty normalized vocabulary. This is how much human can adapt to the new world.
One day when the machine intelligence surpasses human intelligence, we will all speak in Python or whatever the most popular among machines.
