Emotion vector activations across post-training
【启发】情绪向量在后训练阶段的变化轨迹,启发了一个新的训练监控指标体系:目前评估 RLHF 效果主要看 benchmark 分数,但情绪向量的分布变化可能是更敏感的「副作用探测器」——比如,如果某轮 RLHF 意外地使「恐惧」向量激活阈值降低,可能预示着模型在高压场景下更容易产生顺从性偏差。情绪向量或许可以成为训练过程中的「生理指标」。
Emotion vector activations across post-training
【启发】情绪向量在后训练阶段的变化轨迹,启发了一个新的训练监控指标体系:目前评估 RLHF 效果主要看 benchmark 分数,但情绪向量的分布变化可能是更敏感的「副作用探测器」——比如,如果某轮 RLHF 意外地使「恐惧」向量激活阈值降低,可能预示着模型在高压场景下更容易产生顺从性偏差。情绪向量或许可以成为训练过程中的「生理指标」。
Moreover, if processing an event may have external side-effectsbesides updating a replica state – for example, if it may trigger anemail to be sent – then the time warp approach requires some wayof undoing or compensating for those side-effects in the case wherea previously processed event is affected by a late-arriving eventwith an earlier timestamp. It is not possible to un-send an emailonce it has been sent, but it is possible to send a follow-up emailwith a correction, if necessary. If the possibility of such correctionsis unacceptable, optimistic replication cannot be used, and SMR oranother strongly consistent approach must be used instead. In manybusiness systems, corrections or apologies arise from the regularcourse of business anyway [27], so maybe occasional correctionsdue to out-of-order events are also acceptable in practice.
side effects
副作用很关键
that would only be a side effect of its main purpose
main purpose, get rid of annoying errors caused by side effect imports
Michael Bang Petersen. (2021, March 17). This is worsened as costs of #covid19 are not mentally similar to costs of side effects, even if the latter are less risky. People prefer controllable risks to uncontrollabe risks, even if less lethal (https://t.co/kSIcObWYmT). That is why you fear flying but not driving. [2/2] [Tweet]. @M_B_Petersen. https://twitter.com/M_B_Petersen/status/1372103708218159109
Kai Kupferschmidt. (2021, March 16). “I’m not here to give you the outcome of any scientific review”, says EMA director Emer Cooke at start of press conference on AstraZeneca vaccine safety. ‘I’m here to explain the steps in the process, what we’re doing, and when you can expect us to come to a conclusion.’ [Tweet]. @kakape. https://twitter.com/kakape/status/1371811123197001729
Helen Branswell 🇺🇦 on Twitter. (n.d.). Twitter. Retrieved April 1, 2022, from https://twitter.com/HelenBranswell/status/1478835175580741633
Edward Nirenberg 🇺🇦. (2022, March 2). I see a lot of people sharing a document from Pfizer describing pharmacovigilance data from spontaneous reporting and the vast, vast majority of people are not interpreting it correctly so here’s a thread on what it actually says. 🧵 https://phmpt.org/wp-content/uploads/2021/11/5.3.6-postmarketing-experience.pdf [Tweet]. @ENirenberg. https://twitter.com/ENirenberg/status/1498852915758485509
Yang, M. (2022, February 14). Arkansas jail’s ivermectin experiments recall historical medical abuse of imprisoned minorities. The Guardian. https://www.theguardian.com/us-news/2022/feb/14/arkansas-prison-covid-19-ivermectin-experiment-minorities-medical-abuse
Devlin, H., & correspondent, H. D. S. (2022, February 3). Novavax Covid vaccine approved for use in over-18s in UK. The Guardian. https://www.theguardian.com/society/2022/feb/03/novavax-covid-vaccine-approved-for-use-in-over-18s-in-uk
Henley, J. (2022, January 5). Covid ‘vaccination doubt line’ receiving up to 1,000 calls a day in Netherlands. The Guardian. https://www.theguardian.com/world/2022/jan/05/covid-vaccination-doubt-line-receiving-up-to-1000-calls-a-day-in-the-netherlands
Calgary Herald. (n.d.). Calgaryherald. Retrieved December 21, 2021, from https://calgaryherald.com/news/local-news/anti-covid-vaccine-dad-loses-battle-with-ex-wife-over-whether-to-get-their-children-inoculated
Doctor dismantles typical COVID vaccine hesitancy arguments by putting forward alternative scenario. (n.d.). Sky News. Retrieved December 21, 2021, from https://news.sky.com/story/doctor-dismantles-typical-covid-vaccine-hesitancy-arguments-by-putting-forward-alternative-scenario-12501207
Antivaccine activists use a government database on side effects to scare the public. (n.d.). Retrieved December 7, 2021, from https://www.science.org/content/article/antivaccine-activists-use-government-database-side-effects-scare-public
Mahase, E. (2021). Covid-19: Antibody boost after third dose varies greatly by vaccine, study finds. BMJ, 375, n3011. https://doi.org/10.1136/bmj.n3011
Pickles, K., Copp, T., Dodd, R. H., Cvejic, E., Seale, H., Steffens, M. S., Meyerowitz-Katz, G., Bonner, C., & McCaffery, K. (2021). COVID-19 vaccine intentions in Australia. The Lancet Infectious Diseases, 21(12), 1627–1628. https://doi.org/10.1016/S1473-3099(21)00686-1
FDA authorizes Pfizer-BioNTech, Moderna boosters for all adults. (2021). Washington Post. Retrieved November 22, 2021, from https://www.washingtonpost.com/health/2021/11/19/fda-clears-pfizer-biontech-moderna-booster-shots-all-adults/
Claims about changes to Pfizer vaccine for children miss important details. (16:48:08.203282+00:00). Full Fact. https://fullfact.org/online/claims-about-changes-pfizer-vaccine-children-miss-important-details/
Reardon, S. (2021). Cuba’s bet on home-grown COVID vaccines is paying off. Nature. https://doi.org/10.1038/d41586-021-03470-x
Otte, J. (2021, November 11). ‘No jab, no job’: Care home workers in England on the Covid vaccine mandate. The Guardian. https://www.theguardian.com/society/2021/nov/11/england-care-home-workers-on-mandatory-covid-vaccines
Covid-19 vaccination in children, adolescents, and young adults: How can we ensure high vaccination uptake? - The BMJ. (n.d.). Retrieved October 14, 2021, from https://blogs.bmj.com/bmj/2021/10/14/covid-19-vaccination-in-children-adolescents-and-young-adults-how-can-we-ensure-high-vaccination-uptake
Flu and Covid jabs safe to be given at same time, study finds | Vaccines and immunisation | The Guardian. (n.d.). Retrieved October 1, 2021, from https://www.theguardian.com/society/2021/sep/30/flu-covid-jabs-safe-same-time-study?CMP=Share_iOSApp_Other
Stop letting NBA anti-vaxxers spout COVID-19 misinformation. (n.d.). Retrieved September 28, 2021, from https://ca.sports.yahoo.com/news/nba-covid-19-vaccine-kyrie-irving-stop-giving-vocal-minority-space-loud-wrong-023626133.html
Analytics modules that run in the background, monitor user interaction, and send the data to a server.
Many jQuery plugins attach themselves to the global jQuery object.
A polyfill for example, might not do anything, because it finds that the feature that it enables is already supported by the browser.
A module with side-effects is one that changes the scope in other ways then returning something, and it's effects are not always predictable, and can be affected by outside forces (non pure function).
Import a module for its side effects only Import an entire module for side effects only, without importing anything. This runs the module's global code, but doesn't actually import any values.
Telstra on Twitter: “I’m fully aware this won’t stop ‘the vaxx gave me 5G’ jokes. But #LetsVaxx anyway. Https://t.co/IwM7AwrpJ6” / Twitter. (n.d.). Retrieved September 9, 2021, from https://twitter.com/Telstra/status/1433191938497282051
Mel 💉🩹❤️Get Vaccinated! On Twitter: “#GetVaccinatedNow https://t.co/ewb1vQCvWw” / Twitter. (n.d.). Retrieved August 27, 2021, from https://twitter.com/ScientistMel/status/1430927342361137156
How to trash confidence in a COVID-19 vaccine: Brexit edition—Bulletin of the Atomic Scientists. (n.d.). Retrieved August 10, 2021, from https://thebulletin.org/2021/08/how-to-trash-confidence-in-a-covid-19-vaccine-brexit-edition/#.YQwD9u6LazM.twitter
Vaccine-skeptical Tennesseans on finally getting the COVID shot. (n.d.). Retrieved August 4, 2021, from https://slate.com/news-and-politics/2021/07/tennessee-covid-vaccines.html?via=rss_socialflow_twitter
Covid-19: New poll shows why some vaccine-hesitant Americans decided to get the shot—CNNPolitics. (n.d.). Retrieved August 2, 2021, from https://edition.cnn.com/2021/07/13/politics/poll-covid-19-vaccine-decision/index.html
KFF COVID-19 Vaccine Monitor: In Their Own Words, Six Months Later | KFF. (n.d.). Retrieved August 2, 2021, from https://www.kff.org/coronavirus-covid-19/poll-finding/kff-covid-19-vaccine-monitor-in-their-own-words-six-months-later/
Unvaccinated Is Different From Anti-Vax—The Atlantic. (n.d.). Retrieved August 1, 2021, from https://www.theatlantic.com/health/archive/2021/07/unvaccinated-different-anti-vax/619523/
Snopes.com. “No, There Weren’t More COVID-19 Vaccine Deaths Than COVID-19 Deaths.” Accessed July 23, 2021. https://www.snopes.com/fact-check/covid-vax-deaths-vs-covid-deaths/.
has the same effect (that is no side effect)
The difference between PUT and POST is that PUT is idempotent: calling it once or several times successively has the same effect (that is no side effect), whereas successive identical POST requests may have additional effects, akin to placing an order several times.
Gargano, J. W., Wallace, M., Hadler, S. C., Langley, G., Su, J. R., Oster, M. E., Broder, K. R., Gee, J., Weintraub, E., Shimabukuro, T., Scobie, H. M., Moulia, D., Markowitz, L. E., Wharton, M., McNally, V. V., Romero, J. R., Talbot, H. K., Lee, G. M., Daley, M. F., & Oliver, S. E. (2021). Use of mRNA COVID-19 Vaccine After Reports of Myocarditis Among Vaccine Recipients: Update from the Advisory Committee on Immunization Practices — United States, June 2021. MMWR. Morbidity and Mortality Weekly Report, 70(27), 977–982. https://doi.org/10.15585/mmwr.mm7027e2
the Guardian. ‘Israel Says 600 Children given Covid Jab Had No Serious Side-Effects’, 10 March 2021. http://www.theguardian.com/society/2021/mar/10/israel-says-600-children-given-covid-jab-had-no-serious-side-effects.
Jesse O’Shea MD, MSc on Twitter: “Okay Twitter! Here is the new vaccine side effect chart (aka reactogenicity) for FDA submitted COVID19 vaccines vs Shingrix & Flu. J&J’s Ad26.COV2.S has the least side effect profile of the COVID vaccines so far. Https://t.co/MFGzWDqQKZ” / Twitter. (n.d.). Retrieved July 2, 2021, from https://twitter.com/JesseOSheaMD/status/1364645966826070016?s=20
The AstraZeneca Vaccine Blood-Clot Issue Won’t Go Away—The Atlantic. (n.d.). Retrieved July 2, 2021, from https://www.theatlantic.com/health/archive/2021/03/astrazeneca-vaccine-blood-clot-issue-wont-go-away/618451/
Persistent Vaccine Myths | KFF. (n.d.). Retrieved July 2, 2021, from https://www.kff.org/coronavirus-covid-19/perspective/persistent-vaccine-myths/?utm_source=newsletter&utm_medium=email&utm_campaign=newsletter_axiosvitals&stream=top
Dr. Jeff Benyacar on Twitter: “@AlexBerenson ‘Even if a link between myocarditis and the vaccine holds up, the condition is usually mild, requiring treatment only with anti-inflammatory drugs, whereas COVID-19 infection can also cause serious disease and long-term side effects, even in young people.’ https://t.co/3VQprF7bIz” / Twitter. (n.d.). Retrieved July 2, 2021, from https://twitter.com/jbenyacar/status/1399851524487106562?s=12
Subbaraman, N. (2021). This COVID-vaccine designer is tackling vaccine hesitancy—In churches and on Twitter. Nature, 590(7846), 377–377. https://doi.org/10.1038/d41586-021-00338-y
Razai, M. S., Osama, T., McKechnie, D. G. J., & Majeed, A. (2021). Covid-19 vaccine hesitancy among ethnic minority groups. BMJ, n513. https://doi.org/10.1136/bmj.n513
This is Our Shot 🇨🇦 #ThisIsOurShotCA on Twitter: “Tonight is the night! LIVE Town Hall. Real experts. Real answers. 8 pm ET/ 5 pm PT. Tune in at https://t.co/FvdIjdi9SF https://t.co/4ugZ3kTaef” / Twitter. (n.d.). Retrieved June 28, 2021, from https://twitter.com/thisisourshotca/status/1395017500539097095
Michael Makris on Twitter: “The cumulative incidence of VITT after AZ first vaccination in the UK is continuing to increase with the latest data being 1 in 81,000. However this is rather misleading because VITT is age related with a higher incidence in the young. Https://t.co/CSjshAoRsN” / Twitter. (n.d.). Retrieved June 27, 2021, from https://twitter.com/ProfMakris/status/1395457748721184777?s=20
Report a problem or side effect | Therapeutic Goods Administration (TGA). (n.d.). Retrieved June 26, 2021, from https://www.tga.gov.au/reporting-problems
FDA to add warning about rare heart inflammation to Pfizer, Moderna vaccines | Reuters. (n.d.). Retrieved June 25, 2021, from https://www.reuters.com/business/healthcare-pharmaceuticals/us-panel-review-heart-inflammation-cases-after-pfizer-moderna-vaccines-2021-06-23
Imperial College London. (2021, February). Covid-19: Global attitudes towards a COVID-19 vaccine. https://www.imperial.ac.uk/media/imperial-college/institute-of-global-health-innovation/EMBARGOED-0502.-Feb-21-GlobalVaccineInsights_ICL-YouGov-Covid-19-Behaviour-Tracker_20210301.pdf
Remmel, A. (2021). COVID vaccines and safety: What the research says. Nature, 590(7847), 538–540. https://doi.org/10.1038/d41586-021-00290-x
‘No data’ linking Covid vaccines to menstrual changes, US experts say | Coronavirus | The Guardian. (n.d.). Retrieved June 18, 2021, from https://www.theguardian.com/world/2021/apr/23/covid-vaccines-periods-menstruation-changes-data-experts?CMP=Share_iOSApp_Other
Thompson, D. (2021, May 3). Millions Are Saying No to the Vaccines. What Are They Thinking? The Atlantic. https://www.theatlantic.com/ideas/archive/2021/05/the-people-who-wont-get-the-vaccine/618765/
Ian Brown wrong to claim Japan refuses blood from anyone who’s been vaccinated—Full Fact. (n.d.). Retrieved June 5, 2021, from https://fullfact.org/online/ian-brown-wrong-to-claim-japan-refuses-blood-from-anyone-whos-been-vaccinated/
Wood, S., & Schulman, K. (2021). When Vaccine Apathy, Not Hesitancy, Drives Vaccine Disinterest. JAMA. https://doi.org/10.1001/jama.2021.7707
Wadman, M. (2021). Antivaccine activists use a government database on side effects to scare the public. Science. https://doi.org/10.1126/science.abj6981
KFF. “KFF COVID-19 Vaccine Monitor – April 2021,” May 6, 2021. https://www.kff.org/coronavirus-covid-19/poll-finding/kff-covid-19-vaccine-monitor-april-2021/.
the Guardian. “Under-30s in UK Should Be Offered Alternative Covid Vaccine to AstraZeneca Jab, Says Advisory Body,” April 7, 2021. http://www.theguardian.com/society/2021/apr/07/under-30s-in-uk-should-be-offered-alternative-covid-vaccine-to-astrazeneca-jab-says-regulator.
the Guardian. ‘How UK Doctor Linked Rare Blood-Clotting to AstraZeneca Covid Jab’, 13 April 2021. http://www.theguardian.com/society/2021/apr/13/how-uk-doctor-marie-scully-blood-clotting-link-astrazeneca-covid-jab-university-college-london-hospital.
Dr Duncan Robertson. ‘The JCVI/MHRA Recommendation to Restrict Oxford/AZ in under-30s in the UK (Where There Is “low” Exposure Risk) Is consistent with the EMA Recommendation Not to Do so in Europe (Where There Is “Medium” or “High” Exposure Risk) Meaning the Risk/Benefit Balance Changes. Https://T.Co/C6SS9oN3Vz’. Tweet. @Dr_D_Robertson (blog), 7 April 2021. https://twitter.com/Dr_D_Robertson/status/1379808945750085643.
Kai Kupferschmidt. ‘“Our Position Has Not Changed”, Says @EMA_News Head Emer Cooke about AZ Vaccine. “According to the Current Scientific Knowledge, There Is No Evidence That Would Support Restricting the Use of This Vaccine in Any Population."’. Tweet. @kakape (blog), 31 March 2021. https://twitter.com/kakape/status/1377268296739913728.
ReconfigBehSci. ‘@sarahflecke “Reports Emerging of Rare Types of Multiple Thrombosis, Bleeding, and Thrombocytopenia .. Similar to Disseminated Intravasc. Coagulation ... in Otherwise Healthy Individuals Shortly after Receiving ..AstraZeneca ..Vaccine. These Outcomes Are Not Included in the Present Analysis.”’ Tweet. @SciBeh (blog), 2 April 2021. https://twitter.com/SciBeh/status/1377984798422077446.
the Guardian. ‘Small Number of Facebook Users Responsible for Most Covid Vaccine Skepticism – Report’, 16 March 2021. http://www.theguardian.com/technology/2021/mar/15/facebook-study-covid-vaccine-skepticism.
the Guardian. ‘There’s No Proof the Oxford Vaccine Causes Blood Clots. So Why Are People Worried? | David Spiegelhalter’, 15 March 2021. http://www.theguardian.com/commentisfree/2021/mar/15/evidence-oxford-vaccine-blood-clots-data-causal-links.
In production, you will never trigger one specific callback or a particular validation, only. Your application will run all code required to create a Song object, for instance. In Trailblazer, this means running the Song::Create operation, and testing that very operation with all its side-effects.
Responders don't use valid? to check for errors in models to figure out if the request was successful or not, and relies on your controllers to call save or create to trigger the validations.
French professor faces disciplinary case over hydroxychloroquine claims | France | The Guardian. (n.d.). Retrieved March 3, 2021, from https://www.theguardian.com/world/2020/nov/12/covid-professor-didier-raoult-hydroxychloroquine
Patching has no implicit, magical side-effects and is strongly encouraged to customize flows for a specific case in a quick and consise way.
Gee, Julianne. ‘First Month of COVID-19 Vaccine Safety Monitoring — United States, December 14, 2020–January 13, 2021’. MMWR. Morbidity and Mortality Weekly Report 70 (2021). https://doi.org/10.15585/mmwr.mm7008e3.
Also, in non-functional programming, a function without arguments can be meaningful and not necessarily constant (due to side effects).
The IO monad wraps computations in the following context: "This computation can read information from or write information to the terminal, file system, operating system, and/or network". If you want to get user input, print a message to the user, read information from a file, or make a network call, you'll need to do so within the IO Monad. These are "side effects". We cannot perform them from "pure" Haskell code.
I have a Post object that has_one :schedule with accepts_nested_attributes_for :schedule as well. The latter method sets autosave: true, which unfortunately has the effect of hoisting up errors into the parent object so that the errors object on Post looks like this: (byebug) post.errors.details {:"schedule.publish_at"=>[{:error=>:blank}]}
My only concern with this approach is that if someone calls #valid? on the form object afterwards, it would under the hood currently delete the existing errors on the form object and revalidate. The could have unexpected side effects where the errors added by the models passed in or the service called will be lost.
My concern with this approach is still that it's somewhat brittle with the current implementation of valid? because whilst valid? appears to be a predicate and should have no side effects, this is not the case and could remove the errors applied by one of the steps above.
However, sometimes actions can't be rolled back and it is unfortunately unavoidable. For example, consider when we send emails during the call to process. If we send before saving a record and that record fails to save what do we do? We can't unsend that email.
I typically save everything I can first, and then call the side-effects afterwards. If the side-effects fail I can handle them elsewhere and retry when necessary.
Professor Gordon Dougan on Twitter. (n.d.). Twitter. Retrieved 10 February 2021, from https://twitter.com/GordonDougan1/status/1357761529609920518
It's very unusual to use map to instigate side effects like that, you should use forEach for that use case.
If tree-shaking still fails, it's because Rollup thinks that there are side-effects somewhere in your code.
This does solve the problem, but now our project and API is structured differently. In large projects it might be very hard to determine how to pull this trick off, or even impossible!
Instead of using classes and local state, Deku just uses functions and pushes the responsibility of all state management and side-effects onto tools like Redux.
Data fetching, setting up a subscription, and manually changing the DOM in React components are all examples of side effects.
preserve: this preserves all imports whose values are never used. This can cause imports/side-effects to be preserved.
Rather than thinking of useEffect as one function doing the job of 3 separate lifecycles, it might be more helpful to think of it simply as a way to run side effects after render – including the potential cleanup you’d want to do before each one, and before unmounting.
By bringing both designers and non-designers alike into Figma, they create a cross-side network effect. In a direct network effect, a homogenous group gets more value from a product as more of them join. In contrast, a cross-side network effect involves two (or more) distinct groups that grow in size and value as the other group does, too. Figma’s cross-side network effect between designers and non-designers is one of the primary and under-appreciated sources of their compounding success over the last few years. As more designers use Figma, they pull in the non-designers they work with. Similarly, as these non-designers use Figma, they encourage the other designers they work with to use Figma. It’s a virtuous circle and a powerful compounding loop.
By bringing non-designers into the design process, Figma created cross-side network effects for itself.
Where typically the designers would get their designer peers to use the tools they're excited about, now non-designers would experience the value and recommend Figma to designers and non-designers alike.
Anything that is added to a system to improve it (or make it more reliable!) increases complexity, thus uncertainty and risk. We have a bad habit of trying to add “more” to fix a problem, increase the layers of safety, band-aid over a system vulnerability, etc. We don’t often evaluate this, but this added complexity can (and does) make things worse. Wherever possible — and I know it isn’t always possible — remove something isntead of adding it. This is not usually politically favorable, something that Sr. Software Engineers who reduce the codebase size have often heard, but has far fewer side effects.
RESULTS: After adjustment for potential confounders, use of azithromycin (adjusted odds ratio [OR] 1.65, 95% confidence interval [CI] 1.34–2.02; 110 exposed cases), clarithromycin (adjusted OR 2.35, 95% CI 1.90–2.91; 111 exposed cases), metronidazole (adjusted OR 1.70, 95% CI 1.27–2.26; 53 exposed cases), sulfonamides (adjusted OR 2.01, 95% CI 1.36–2.97; 30 exposed cases), tetracyclines (adjusted OR 2.59, 95% CI 1.97–3.41; 67 exposed cases) and quinolones (adjusted OR 2.72, 95% CI 2.27–3.27; 160 exposed cases) was associated with an increased risk of spontaneous abortion.
This shows between a 65% to 272% increased risk of miscarriage from use of common antibiotics.
The risk is being characterized as "small' - http://www.cnn.com/2017/05/01/health/antibiotics-miscarriage-study/
I'm not sure any pregnant woman would characterize more than doubling of the risk as small, and we can assume that if this data was shown for a natural substance, it would be characterized differently.
Credibull score = 0.33 / 10
To provide feedback on the score fill in the form available here
What is Credibull? getcredibull.com