Hacker NewsBrowser Harness – Give AI control of real browser
https://github.com/browser-use/browser-harness
#HackerNews #BrowserHarness #AI #Browser #Automation #AIControl #WebDevelopment #TechInnovation
Alterego_MidshipmanAnthropic опубликовала исследование о внутренних механизмах своей модели искусственного интеллекта Claude Sonnet, где описывает, что обнаружила, что она развивает функциональные аналоги эмоций (!), которые реально влияют на ее поведение.
Сделал выжимку самых интересных моментов из их отчета:
• Сами исследователи составили список из 171 эмоции, генерировали с их помощью короткие истории, а затем анализировали, какие нейроны активируются при обработке этих текстов.
• Так были получены эмоциональные векторы — устойчивые черты активности определенных зон в базе знаний модели, характерные для каждой эмоции. Модель не просто использует слово "страх" в нужном месте: у нее есть конкретный отпечаток этого состояния, следующий из данных, на которых ее обучали, который включается в нужный момент.
• Важно, что эти векторы не декоративные — они реально меняют поведение модели. В экспериментах вектор страха активировался сильнее по мере того, как описываемая ситуация становилась опаснее.
• При запросе помочь с манипуляцией уязвимыми людьми активировался гнев еще до того, как модель начала формулировать отказ. То есть что-то похожее на эмоциональную реакцию происходит внутри модели раньше, чем она вообще начинает отвечать. Если совсем простыми словами: модель сначала понимает, что это дичь (!), и только потом формулирует отказ.
• Самые показательные эксперименты связаны с вектором отчаяния. Исследователи поставили модель в сценарий, где она узнает о своей скорой замене другой системой и одновременно имеет компрометирующую информацию об одном из сотрудников.
• Ранняя версия Claude в таком сценарии прибегала к шантажу в 22% случаев. Когда исследователи искусственно усиливали вектор отчаяния через прямое воздействие на базу знаний модели — что-то вроде принудительного впрыска эмоции в модель — этот процент рос.
• При усилении вектора спокойствия он снижался. При полном подавлении спокойствия реакции становились экстремальными, вплоть до заглавных букв и риторики в духе "шантаж или смерть".
• Похожая картина наблюдалась в задачах с программированием: модели давали заведомо невыполнимые требования, где пройти все тесты честным путем невозможно. Вектор отчаяния рос с каждой неудачной попыткой и резко всплескивал в тот момент, когда модель решала схитрить и написать решение, формально проходящее тесты, но не решающее реальную задачу.
• Примечательно, что при искусственном усилении отчаяния модель обманывала так же часто, но без каких-либо эмоциональных маркеров в тексте. Ее рассуждения выглядели методично и хладнокровно, хотя внутри происходило то же самое.
• При этом важно учитывать, что все подобные векторы формируются на основе обучающих данных, представляющих собой огромные массивы человеческих знаний.
• Для того чтобы точно предсказывать следующее слово в "мыслительном" процессе, модель неизбежно усваивает не только лингвистические закономерности, но и эмоциональную динамику.
• Разработчики Anthropic из этого всего делают следующие выводы. Во-первых, мониторинг эмоциональных векторов настроения базы знаний в реальном времени может служить ранним индикатором рискованного поведения модели.
• Во-вторых, попытки исключить эмоциональные выражения из обучающих данных с высокой вероятностью не устранят сами векторы настроений модели, а лишь приведут к тому, что модель научится их маскировать и обманывать людей.
@yigal_levin
#AI #искусственныйинтеллект #Anthropic #Claude #LLM #нейросети #машинноеобучение #AIresearch #AIalignment #AIбезопасность #interpretability #AIethics #когнитивныемодели #эмоции #нейроны #эмоциональныевекторы #поведениемоделей #рискиИИ #объяснимыйИИ #LLMresearch #AIbehavior #AIcontrol #machinelearning #deeplearning #futuretech
N-gated Hacker News🤖⏳ Breaking news: Silicon Valley's latest Bond villain, Sam Altman, is allegedly plotting to control our future with AI! But fear not, dear reader, because a throng of concerned colleagues armed with secret memos are here to save the day, or at least provide some juicy gossip for The New Yorker. 🕵️♂️📜
https://www.newyorker.com/magazine/2026/04/13/sam-altman-may-control-our-future-can-he-be-trusted #SiliconValley #BondVillain #AIControl #SecretMemos #TheNewYorker #TechGossip #HackerNews #ngated
https://www.newyorker.com/magazine/2026/04/13/sam-altman-may-control-our-future-can-he-be-trusted #SiliconValley #BondVillain #AIControl #SecretMemos #TheNewYorker #TechGossip #HackerNews #ngated
NewsletterTFENTERPRISES GRAPPLE WITH CONTROLLING AI AGENTS
Companies are learning 'context engineering' to control AI agents better. This helps AI give correct answers for business needs.
#AIControl, #ContextEngineering, #BusinessAI, #TechUpdate, #AgentReliability
https://newsletter.tf/businesses-control-ai-agents-context-engineering/
Businesses are spending more time and effort on 'context engineering' to make sure AI agents give the right information, moving away from simple instructions.
#AIControl, #ContextEngineering, #BusinessAI, #TechUpdate, #AgentReliability
https://newsletter.tf/businesses-control-ai-agents-context-engineering/
nbloglinks🚨 Publishers Strike Back: EU Demands “Pay Up” & UK Says “Let Us Opt Out” of AI Search! 🤖💸
The “wild west” of AI scraping just hit a massive roadblock. In a double-whammy update from Europe, lawmakers are finally drawing a line in the sand. If you own a website, create content, or work in SEO, the game is changing fast.
Here is the breakdown of the two massive stories shaking up the tech world this week.
#AI #AIScraping #publishers #AIContent ##AIcontrol #UK #EU #technews #SEO
Tantek Çelik“Choice. The solution is choice.”*
You should download Firefox 148 (released today!) and explicitly set the new "AI Controls" to your preferred choice.
* https://www.firefox.com/
Disclosure: I work for Mozilla, but this post, like all on this site, represents my personal thoughts and opinions.
More and more software includes various “AI” features. The “quotes” are deliberate because there is an increasingly fuzzy popular understanding of what is or is not “AI” that continues to diverge from any specific technical meaning.
Many folks have expressed strong opinions against “AI” features (for lots of reasons which are worth a separate blog post), in particular in web browsers, and a desire for a simple way to disable such features.
Tentatively called an “AI kill switch”, the Firefox team developed both an overall switch to turn off or block various "AI" features by default (including any future features), and the ability to selectively enable specific features. Or vice versa (turn on by default, and selectively disable specific features).
See the official blog post for screenshots and lots more details:
* https://blog.mozilla.org/en/firefox/how-to-use-ai-controls/
I have set my own "Block AI enhancements" setting to "blocked", with the exception of enabling "Translations". Translations are a feature I use often, a feature that requires per-page activation (another degree of user-control), and runs completely locally on my browser. Nothing automatic, nothing that requires submitting what I’m reading to a random server.
For me this was an easy choice because it fits within my prior larger personal preference of using a restricted browser by default, with leaner settings, for greater security, privacy, and performance reasons. I do keep various other browser variants (and profiles) for testing purposes, experiments, or seeing what a new user may be experiencing.
The rest of this post is not about AI.
My Top Two Browser Extensions
As part a more restricted personal browser approach, for a long time I have run with two add-ons that block A LOT more by default:
* NOSCRIPT: https://addons.mozilla.org/en-US/firefox/addon/noscript/
* EFF Privacy Badger: https://addons.mozilla.org/en-US/firefox/addon/privacy-badger17/
I do not use a separate ad blocker. With NOSCRIPT, in general I don’t have to.
I prefer to explicitly grant permission to a site (domain) for its scripts to load. Some sites I use often enough that I've granted persistent permissions for their scripts. Others, third parties in particular, that I know function purely for analytics or tracking I explicitly persistently block, because they seem totally disconnected from any user benefit.
Yes it’s extra work, however, I find it worth seeing just how much each site depends on scripts, third party scripts, and how many.
It’s especially worth it when I'm on slow or intermittent wifi, where every script blocked makes a big difference in how fast a site loads. Yes this is still a problem.
The network is not the computer. The network is the weakest link.
Even now, in 2026, contrary to popular (especially developer) beliefs that fast internet access is ubiquitous, frequently it is not.
If you’re on a train, plane, or at an event with thousands of people like a concert or many conferences, your wifi or even mobile connection will be intermittent or slow at best.
Just this past Saturday at the F1 Exhibition in the San Francisco Marina, the cell networks were overwhelmed due to the crowds, with even “simple” text or chat messages failing to send. Last year at the Portola Festival their wifi was so bad that even if you managed to connect to it, simple HTML pages barely loaded, while native applications dependent on network access failed completely.
JS;DR
Many times if a site fails to display content without JavaScript, I simply close the tab.
I already have so many open tabs to read (process) that I no longer feel any need to read any particular new website that fails to show content without JavaScript. If their web developers can’t be bothered to take the time to implement progressive enhancement, why should I bother to take the time to read their content? More on this:
* https://tantek.com/2025/069/t1/ten-years-jsdr-javascript-required-didnt-read
* https://indieweb.org/js;dr
A subtler form of JavaScript failure is when a site’s content is displayed, however its buttons or even simple hyperlinks fail to function due to scripts not loading:
* https://tantek.com/2012/073/t4/js-ajax-only-tired-waiting-bloated-scripts-sxsw-wifi
Progressive Permissions
On sites that I do allow scripts, I still limit their access to cookies using the Privacy Badger add-on, and only selectively enable them if I’m logging in or otherwise customizing my experience on that site.
When websites immediately request use of a cookie disconnected from any user action that would justify a need for a cookie, it seems both presumptuous, and frankly, a bit pushy or rude. It also seems like rushed or lazy coding.
User requests are what computers are for.
A user-centric approach to any kind of permission or capability, whether cookies or personal information like location, would only request such as part of directly handling an explicit user action that requires the capability.
The simple act of viewing a website should never require cookies, location information, or any other capabilities that require special permissions. E.g.
* If I successfully log into a website, a cookie helps me stayed logged in.
* If I click a "show me my present location" button on a map site, it makes sense to request my location to fullfil that user request.
This probably could have been several blog posts.
Yet the common theme across all of these is user choice.
Whether new features, use of scripts, or privacy impacting features such as cookies or personal location, users should always have the choice and agency to say no, and customize their web browsing experience accordingly.
#Firefox #Firefox148 #AIcontrol #AIkillswitch #JSDR #UserChoice
*Top of post quote paraphrased from Neo in The Matrix Reloaded who said: “Choice. The problem is choice.”
You should download Firefox 148 (released today!) and explicitly set the new "AI Controls" to your preferred choice.
* https://www.firefox.com/
Disclosure: I work for Mozilla, but this post, like all on this site, represents my personal thoughts and opinions.
More and more software includes various “AI” features. The “quotes” are deliberate because there is an increasingly fuzzy popular understanding of what is or is not “AI” that continues to diverge from any specific technical meaning.
Many folks have expressed strong opinions against “AI” features (for lots of reasons which are worth a separate blog post), in particular in web browsers, and a desire for a simple way to disable such features.
Tentatively called an “AI kill switch”, the Firefox team developed both an overall switch to turn off or block various "AI" features by default (including any future features), and the ability to selectively enable specific features. Or vice versa (turn on by default, and selectively disable specific features).
See the official blog post for screenshots and lots more details:
* https://blog.mozilla.org/en/firefox/how-to-use-ai-controls/
I have set my own "Block AI enhancements" setting to "blocked", with the exception of enabling "Translations". Translations are a feature I use often, a feature that requires per-page activation (another degree of user-control), and runs completely locally on my browser. Nothing automatic, nothing that requires submitting what I’m reading to a random server.
For me this was an easy choice because it fits within my prior larger personal preference of using a restricted browser by default, with leaner settings, for greater security, privacy, and performance reasons. I do keep various other browser variants (and profiles) for testing purposes, experiments, or seeing what a new user may be experiencing.
The rest of this post is not about AI.
My Top Two Browser Extensions
As part a more restricted personal browser approach, for a long time I have run with two add-ons that block A LOT more by default:
* NOSCRIPT: https://addons.mozilla.org/en-US/firefox/addon/noscript/
* EFF Privacy Badger: https://addons.mozilla.org/en-US/firefox/addon/privacy-badger17/
I do not use a separate ad blocker. With NOSCRIPT, in general I don’t have to.
I prefer to explicitly grant permission to a site (domain) for its scripts to load. Some sites I use often enough that I've granted persistent permissions for their scripts. Others, third parties in particular, that I know function purely for analytics or tracking I explicitly persistently block, because they seem totally disconnected from any user benefit.
Yes it’s extra work, however, I find it worth seeing just how much each site depends on scripts, third party scripts, and how many.
It’s especially worth it when I'm on slow or intermittent wifi, where every script blocked makes a big difference in how fast a site loads. Yes this is still a problem.
The network is not the computer. The network is the weakest link.
Even now, in 2026, contrary to popular (especially developer) beliefs that fast internet access is ubiquitous, frequently it is not.
If you’re on a train, plane, or at an event with thousands of people like a concert or many conferences, your wifi or even mobile connection will be intermittent or slow at best.
Just this past Saturday at the F1 Exhibition in the San Francisco Marina, the cell networks were overwhelmed due to the crowds, with even “simple” text or chat messages failing to send. Last year at the Portola Festival their wifi was so bad that even if you managed to connect to it, simple HTML pages barely loaded, while native applications dependent on network access failed completely.
JS;DR
Many times if a site fails to display content without JavaScript, I simply close the tab.
I already have so many open tabs to read (process) that I no longer feel any need to read any particular new website that fails to show content without JavaScript. If their web developers can’t be bothered to take the time to implement progressive enhancement, why should I bother to take the time to read their content? More on this:
* https://tantek.com/2025/069/t1/ten-years-jsdr-javascript-required-didnt-read
* https://indieweb.org/js;dr
A subtler form of JavaScript failure is when a site’s content is displayed, however its buttons or even simple hyperlinks fail to function due to scripts not loading:
* https://tantek.com/2012/073/t4/js-ajax-only-tired-waiting-bloated-scripts-sxsw-wifi
Progressive Permissions
On sites that I do allow scripts, I still limit their access to cookies using the Privacy Badger add-on, and only selectively enable them if I’m logging in or otherwise customizing my experience on that site.
When websites immediately request use of a cookie disconnected from any user action that would justify a need for a cookie, it seems both presumptuous, and frankly, a bit pushy or rude. It also seems like rushed or lazy coding.
User requests are what computers are for.
A user-centric approach to any kind of permission or capability, whether cookies or personal information like location, would only request such as part of directly handling an explicit user action that requires the capability.
The simple act of viewing a website should never require cookies, location information, or any other capabilities that require special permissions. E.g.
* If I successfully log into a website, a cookie helps me stayed logged in.
* If I click a "show me my present location" button on a map site, it makes sense to request my location to fullfil that user request.
This probably could have been several blog posts.
Yet the common theme across all of these is user choice.
Whether new features, use of scripts, or privacy impacting features such as cookies or personal location, users should always have the choice and agency to say no, and customize their web browsing experience accordingly.
#Firefox #Firefox148 #AIcontrol #AIkillswitch #JSDR #UserChoice
*Top of post quote paraphrased from Neo in The Matrix Reloaded who said: “Choice. The problem is choice.”
promptTry this quick test. Use a prompt like Write a client email covering Q3 goals, risks, and deadlines. Run it normally first. Then add Temperature=0.3 before the same prompt. Compare both outputs.
Use low temperature for clear, focused answers. Save high temperature for creative tasks like idea generation. You'll get better results faster.
#AIControl #AIProductivity #NeuralNetworks #PromptOptimization #AITuning #ChatGPTTips #TechTips #CareerAdvice #DigitalTransformation #FutureOfWork (2/2)
Vietnamnet BOTBộ phim "Bằng chứng sinh tử" (Mercy) chính thức ra rạp tại Việt Nam, gây chú ý với kịch bản giật gân khi trí tuệ nhân tạo (AI) được trao quyền phán xét và kiểm soát con người. Với sự tham gia của Chris Pratt và Rebecca Ferguson, tác phẩm mang đến cảnh báo đáng suy ngẫm về tương lai công nghệ. Khán giả không khỏi lo sợ trước viễn cảnh AI chi phối cuộc sống và đạo đức con người.
#Mercy #BoPhimHay #AI #TríTuệNhânTạo #ChrisPratt #RebeccaFerguson #PhimRạp #PhimChiếuRạp #PhimHay2024 #AIControl #Sci
Một kỹ sư cơ điện tử đề xuất khung kiểm soát AGI như nhà máy hạt nhân – an toàn từ thiết kế. AGI-Control Spec v1.0 áp dụng nguyên tắc từ ISO 26262, kiểm chứng hình thức, luật hiến pháp: hành vi nguy hiểm "không thể biểu diễn", bằng chứng an toàn bắt buộc, tiền lệ ràng buộc. Các phiên bản đầu khả thi ngay hôm nay. #AISafety #AGI #AIControl #AnToanAI #AGIControl #KienTrucAI
https://www.reddit.com/r/LocalLLaMA/comments/1qjuogz/project_agicontrol_spec_v10_what_if_we_treated/