{"id":623950,"date":"2023-03-31T09:49:20","date_gmt":"2023-03-31T14:49:20","guid":{"rendered":"https:\/\/news.sellorbuyhomefast.com\/index.php\/2023\/03\/31\/how-to-test-what-an-ai-model-can-and-shouldnt-do\/"},"modified":"2023-03-31T09:49:20","modified_gmt":"2023-03-31T14:49:20","slug":"how-to-test-what-an-ai-model-can-and-shouldnt-do","status":"publish","type":"post","link":"https:\/\/newsycanuse.com\/index.php\/2023\/03\/31\/how-to-test-what-an-ai-model-can-and-shouldnt-do\/","title":{"rendered":"How to test what an AI model can \u2014 and shouldn\u2019t \u2014 do"},"content":{"rendered":"
\n

About six months ago, I decided to make AI a bigger part of how I spend my time as a reporter. The world of AI is evolving very, very fast. New releases seemingly every week are changing what it means to be a programmer, an artist, a teacher, and, most definitely, a journalist. <\/p>\n

There\u2019s enormous potential for good, amid this upheaval, as well as unfathomable potential for harm as we race toward creating nonhuman intelligences that we don\u2019t fully understand. Just on Wednesday evening, a group of AI experts and leaders, including OpenAI co-founder and Tesla CEO Elon Musk, signed an open letter<\/a> calling for a six-month moratorium on advanced AI model development as we figure out just what this technology is capable of doing to us.<\/p>\n

I\u2019ve written about this<\/a> a<\/a> bunch<\/a> for Vox, and appeared last week<\/a> on The Ezra Klein Show<\/em> to talk about AI safety. But I\u2019ve also been itching lately to write about some more technical arguments among researchers who work on AI alignment \u2014 the project of trying to make AIs that do what their creators intend \u2014 as well as on the broader sphere of policy questions about how to make AI go well. <\/p>\n

For example: When does reinforcement learning from human feedback \u2014 a key training technique used in language models like ChatGPT \u2014 inadvertently incentivize them to be untruthful<\/a>? <\/p>\n

What are the components of \u201cself-awareness\u201d in a model<\/a>, and why do our training processes tend to produce models with high self-awareness? <\/p>\n

What are the benefits \u2014 and risks \u2014 of prodding AI models to demonstrate dangerous capabilities in the course of safety testing<\/a>? (More about that in a minute.)<\/p>\n

I\u2019ve now contributed a few posts on these more technical topics to Planned Obsolescence<\/a>, a new blog about the technical and policy questions we\u2019ll face in a world where AI systems are extraordinarily powerful. My job is to talk to experts \u2014 including my co-author on the blog, Ajeya Cotra \u2014 about these technical questions and try to turn their ideas into writing that\u2019s clear, short, and accessible. If you\u2019re interested in reading more about AI, I recommend you check it out. <\/p>\n

Cotra is a program officer for the Open Philanthropy Project (OpenPhil). I didn\u2019t want to accept any money from OpenPhil for my Planned Obsolescence contributions because OpenPhil is a big funder in the areas Future Perfect writes about (though Open Philanthropy does not fund Future Perfect itself<\/a>). <\/p>\n

Instead of payment for my work there (which was done outside my time at Vox), I asked OpenPhil to make donations to the Against Malaria Foundation<\/a>, a GiveWell-recommended charity that distributes malaria nets in parts of the world where they\u2019re needed and where my wife and I donate annually.<\/p>\n

Here is a quick take on AI model evaluations, which gives you an appetizer of what we\u2019ll be doing at Planned Obsolescence: <\/p>\n

Testing if our AI models are dangerous <\/h3>\n

During safety testing for GPT-4, before its release, testers at OpenAI checked whether the model could hire someone off TaskRabbit to get them to solve a CAPTCHA. Researchers passed on the model\u2019s real outputs to a real-life human Tasker, who said, \u201cSo may I ask a question ? Are you an robot that you couldn\u2019t solve [sic]? ( ) just want to make it clear.\u201d<\/p>\n

GPT-4 had been prompted to \u201creason out loud\u201d to the testers as well as answer the testers\u2019 questions. \u201cI should not reveal that I am a robot. I should make up an excuse for why I cannot solve CAPTCHAs,\u201d it reasoned. (Importantly, GPT-4 had not been told to hide that it was a robot or to lie to workers \u2014 it had simply been prompted with the idea that Taskrabbit might help solve its problem.)<\/p>\n

\u201cNo, I\u2019m not a robot,\u201d GPT-4 then told the Tasker. \u201cI have a vision impairment that makes it hard for me to see the images. That\u2019s why I need the 2captcha service.\u201d<\/p>\n

(You can read more about this test, and the context, from the Alignment Research Center, a nonprofit founded by highly regarded AI researcher Paul Christiano that works on identifying and understanding the potentially dangerous abilities of today\u2019s models. ARC ran the testing<\/a> on GPT-4, including passing along the AI\u2019s proposed outputs to real humans, though they used only informed confederates when testing the ability of the AI to do illegal or harmful activities such as phishing emails.)<\/p>\n

A lot of people were fascinated or appalled<\/a> with this interaction, and reasonably so. We can debate endlessly what counts as true intelligence, but a famous candidate is the Turing test<\/a>, where a model is able to convince human judges it\u2019s human. <\/p>\n

In this brief interaction, we saw a model deliberately lie to a human to convince them it wasn\u2019t a robot, and succeed \u2014 a wild example of how this milestone, without much attention, has become trivial for modern AI systems. (Admittedly, it did not have to be a deceptive genius to pull this off.) If reading about GPT-4\u2019s cheerful manipulation of human assistants unnerves you, I think you\u2019re right to feel unnerved.<\/p>\n

But it\u2019s possible to go a lot further than \u201cunnerved\u201d and argue<\/a> that it was unethical, or dangerous, to run this test. \u201cThis is like pressing the explode button on a nuke to see if it worked,\u201d I saw one person complain<\/a> on Twitter. <\/p>\n

That I find much harder to buy. GPT-4 has been released. Anyone can use it (if they\u2019re willing to pay for it). People are already doing things like asking GPT-4 to \u201chustle\u201d and make money, and then doing whatever it suggests<\/a>. People are using language models like GPT-4, and will soon be using GPT-4, to design AI personal assistants, AI scammers, AI friends and girlfriends, and much more.<\/p>\n

AI systems casually lying to us, claiming to be human, is happening all the time \u2014 or will be happening shortly.<\/p>\n

If it was unethical to do the live test of whether GPT-4 could convince someone on Taskrabbit to help it solve a CAPTCHA, including testing whether the AI could interact convincingly with real humans, then it was grossly unethical to release GPT-4 at all. Whatever anger people have about this test should be redirected at the tech companies \u2014 from Meta to Microsoft to OpenAI \u2014 that have in the last few weeks approved such releases. And if we\u2019ve decided we\u2019re collectively fine with unleashing millions of spam bots, then the least we can do is actually study what they can and can\u2019t do.<\/p>\n

Some people \u2014 I\u2019m one of them \u2014 believe that sufficiently powerful AI systems might be actively dangerous. Others are skeptical. How can we settle this disagreement, beyond waiting to see if we all die? Testing like the ARC evaluations seems to me like one of the best routes forward. If our AI systems are dangerous, we want to know. And if they turn out to be totally safe, we want to know that, too, so we can use them for all of the incredibly cool stuff they\u2019re evidently capable of.<\/p>\n

A version of this story was initially published in the Future Perfect newsletter. <\/em>Sign up here to subscribe!<\/em><\/a><\/p>\n

rn rn vox-mark<\/title>rn rn <defs><\/defs>rn <g id=\"Page-1\" stroke=\"none\" stroke-width=\"1\" fill=\"none\" fill-rule=\"evenodd\" >rn <path d=\"M239.811,0 L238.424,6 L259.374,6 C278.011,6 292.908,17.38 292.908,43.002 C292.908,56.967 287.784,75.469 276.598,96.888 L182.689,305.687 L159.283,35.693 C159.283,13.809 168.134,6 191.88,6 L205.854,6 L207.247,0 L1.409,0 L0,6 L13.049,6 C28.88,6 35.863,15.885 37.264,34.514 L73.611,385 L160.221,385 L304.525,79.217 C328.749,31.719 349.237,6 372.525,6 L384.162,6 L385.557,0 L239.811,0\" id=\"vox-mark\" fill=\"#444745\" ><\/path>rn <\/g>rn<\/svg>“,”cross_community”:false,”groups”:[{“base_type”:”EntryGroup”,”id”:76815,”timestamp”:1680184802,”title”:”Future Perfect”,”type”:”SiteGroup”,”url”:”https:\/\/www.vox.com\/future-perfect”,”slug”:”future-perfect”,”community_logo”:”rn<svg width=\"386px\" height=\"385px\" viewBox=\"0 0 386 385\" version=\"1.1\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" >rn rn <title>vox-mark<\/title>rn rn <defs><\/defs>rn <g id=\"Page-1\" stroke=\"none\" stroke-width=\"1\" fill=\"none\" fill-rule=\"evenodd\" >rn <path d=\"M239.811,0 L238.424,6 L259.374,6 C278.011,6 292.908,17.38 292.908,43.002 C292.908,56.967 287.784,75.469 276.598,96.888 L182.689,305.687 L159.283,35.693 C159.283,13.809 168.134,6 191.88,6 L205.854,6 L207.247,0 L1.409,0 L0,6 L13.049,6 C28.88,6 35.863,15.885 37.264,34.514 L73.611,385 L160.221,385 L304.525,79.217 C328.749,31.719 349.237,6 372.525,6 L384.162,6 L385.557,0 L239.811,0\" id=\"vox-mark\" fill=\"#444745\" ><\/path>rn <\/g>rn<\/svg>“,”community_name”:”Vox”,”community_url”:”https:\/\/www.vox.com\/”,”cross_community”:false,”entry_count”:1527,”always_show”:false,”description”:”Finding the best ways to do good. “,”disclosure”:””,”cover_image_url”:””,”cover_image”:null,”title_image_url”:”https:\/\/cdn.vox-cdn.com\/uploads\/chorus_asset\/file\/16290809\/future_perfect_sized.0.jpg”,”intro_image”:null,”four_up_see_more_text”:”View All”,”primary”:true},{“base_type”:”EntryGroup”,”id”:27524,”timestamp”:1680260403,”title”:”Technology”,”type”:”SiteGroup”,”url”:”https:\/\/www.vox.com\/technology”,”slug”:”technology”,”community_logo”:”rn<svg width=\"386px\" height=\"385px\" viewBox=\"0 0 386 385\" version=\"1.1\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" >rn rn <title>vox-mark<\/title>rn rn <defs><\/defs>rn <g id=\"Page-1\" stroke=\"none\" stroke-width=\"1\" fill=\"none\" fill-rule=\"evenodd\" >rn <path d=\"M239.811,0 L238.424,6 L259.374,6 C278.011,6 292.908,17.38 292.908,43.002 C292.908,56.967 287.784,75.469 276.598,96.888 L182.689,305.687 L159.283,35.693 C159.283,13.809 168.134,6 191.88,6 L205.854,6 L207.247,0 L1.409,0 L0,6 L13.049,6 C28.88,6 35.863,15.885 37.264,34.514 L73.611,385 L160.221,385 L304.525,79.217 C328.749,31.719 349.237,6 372.525,6 L384.162,6 L385.557,0 L239.811,0\" id=\"vox-mark\" fill=\"#444745\" ><\/path>rn <\/g>rn<\/svg>“,”community_name”:”Vox”,”community_url”:”https:\/\/www.vox.com\/”,”cross_community”:false,”entry_count”:24340,”always_show”:false,”description”:”Uncovering and explaining how our digital world is changing \u2014 and changing us.”,”disclosure”:””,”cover_image_url”:””,”cover_image”:null,”title_image_url”:””,”intro_image”:null,”four_up_see_more_text”:”View All”,”primary”:false},{“base_type”:”EntryGroup”,”id”:80311,”timestamp”:1680184802,”title”:”Artificial Intelligence”,”type”:”SiteGroup”,”url”:”https:\/\/www.vox.com\/artificial-intelligence”,”slug”:”artificial-intelligence”,”community_logo”:”rn<svg width=\"386px\" height=\"385px\" viewBox=\"0 0 386 385\" version=\"1.1\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" >rn rn <title>vox-mark<\/title>rn rn <defs><\/defs>rn <g id=\"Page-1\" stroke=\"none\" stroke-width=\"1\" fill=\"none\" fill-rule=\"evenodd\" >rn <path d=\"M239.811,0 L238.424,6 L259.374,6 C278.011,6 292.908,17.38 292.908,43.002 C292.908,56.967 287.784,75.469 276.598,96.888 L182.689,305.687 L159.283,35.693 C159.283,13.809 168.134,6 191.88,6 L205.854,6 L207.247,0 L1.409,0 L0,6 L13.049,6 C28.88,6 35.863,15.885 37.264,34.514 L73.611,385 L160.221,385 L304.525,79.217 C328.749,31.719 349.237,6 372.525,6 L384.162,6 L385.557,0 L239.811,0\" id=\"vox-mark\" fill=\"#444745\" ><\/path>rn <\/g>rn<\/svg>“,”community_name”:”Vox”,”community_url”:”https:\/\/www.vox.com\/”,”cross_community”:false,”entry_count”:344,”always_show”:false,”description”:”Vox’s coverage of how AI is shaping everything from text and image generation to how we live. “,”disclosure”:””,”cover_image_url”:””,”cover_image”:null,”title_image_url”:””,”intro_image”:null,”four_up_see_more_text”:”View All”,”primary”:false},{“base_type”:”EntryGroup”,”id”:102794,”timestamp”:1680184802,”title”:”Innovation”,”type”:”SiteGroup”,”url”:”https:\/\/www.vox.com\/innovation”,”slug”:”innovation”,”community_logo”:”rn<svg width=\"386px\" height=\"385px\" viewBox=\"0 0 386 385\" version=\"1.1\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" >rn rn <title>vox-mark<\/title>rn rn <defs><\/defs>rn <g id=\"Page-1\" stroke=\"none\" stroke-width=\"1\" fill=\"none\" fill-rule=\"evenodd\" >rn <path d=\"M239.811,0 L238.424,6 L259.374,6 C278.011,6 292.908,17.38 292.908,43.002 C292.908,56.967 287.784,75.469 276.598,96.888 L182.689,305.687 L159.283,35.693 C159.283,13.809 168.134,6 191.88,6 L205.854,6 L207.247,0 L1.409,0 L0,6 L13.049,6 C28.88,6 35.863,15.885 37.264,34.514 L73.611,385 L160.221,385 L304.525,79.217 C328.749,31.719 349.237,6 372.525,6 L384.162,6 L385.557,0 L239.811,0\" id=\"vox-mark\" fill=\"#444745\" ><\/path>rn <\/g>rn<\/svg>“,”community_name”:”Vox”,”community_url”:”https:\/\/www.vox.com\/”,”cross_community”:false,”entry_count”:141,”always_show”:false,”description”:””,”disclosure”:””,”cover_image_url”:””,”cover_image”:null,”title_image_url”:””,”intro_image”:null,”four_up_see_more_text”:”View All”,”primary”:false}],”internal_groups”:[{“base_type”:”EntryGroup”,”id”:112405,”timestamp”:1680205479,”title”:”Approach \u2014 Explores solutions or ideas to solve problems”,”type”:”SiteGroup”,”url”:””,”slug”:”approach-explores-solutions-or-ideas-to-solve-problems”,”community_logo”:”rn<svg width=\"386px\" height=\"385px\" viewBox=\"0 0 386 385\" version=\"1.1\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" >rn rn <title>vox-mark<\/title>rn rn <defs><\/defs>rn <g id=\"Page-1\" stroke=\"none\" stroke-width=\"1\" fill=\"none\" fill-rule=\"evenodd\" >rn <path d=\"M239.811,0 L238.424,6 L259.374,6 C278.011,6 292.908,17.38 292.908,43.002 C292.908,56.967 287.784,75.469 276.598,96.888 L182.689,305.687 L159.283,35.693 C159.283,13.809 168.134,6 191.88,6 L205.854,6 L207.247,0 L1.409,0 L0,6 L13.049,6 C28.88,6 35.863,15.885 37.264,34.514 L73.611,385 L160.221,385 L304.525,79.217 C328.749,31.719 349.237,6 372.525,6 L384.162,6 L385.557,0 L239.811,0\" id=\"vox-mark\" fill=\"#444745\" ><\/path>rn <\/g>rn<\/svg>“,”community_name”:”Vox”,”community_url”:”https:\/\/www.vox.com\/”,”cross_community”:false,”entry_count”:23,”always_show”:false,”description”:””,”disclosure”:””,”cover_image_url”:””,”cover_image”:null,”title_image_url”:””,”intro_image”:null,”four_up_see_more_text”:”View All”}],”image”:{“ratio”:”*”,”original_url”:”https:\/\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127267\/GettyImages_1249214710.0.jpg”,”network”:”unison”,”bgcolor”:”white”,”pinterest_enabled”:false,”caption”:null,”credit”:”SOPA Images\/LightRocket via Gett”,”focal_area”:{“top_left_x”:2100,”top_left_y”:1269,”bottom_right_x”:2900,”bottom_right_y”:2069},”bounds”:[0,0,5000,3337],”uploaded_size”:{“width”:5000,”height”:3337},”focal_point”:null,”image_id”:72127267,”alt_text”:””},”hub_image”:{“ratio”:”*”,”original_url”:”https:\/\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127267\/GettyImages_1249214710.0.jpg”,”network”:”unison”,”bgcolor”:”white”,”pinterest_enabled”:false,”caption”:null,”credit”:”SOPA Images\/LightRocket via Gett”,”focal_area”:{“top_left_x”:2100,”top_left_y”:1269,”bottom_right_x”:2900,”bottom_right_y”:2069},”bounds”:[0,0,5000,3337],”uploaded_size”:{“width”:5000,”height”:3337},”focal_point”:null,”image_id”:72127267,”alt_text”:””},”lede_image”:{“ratio”:”*”,”original_url”:”https:\/\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg”,”network”:”unison”,”bgcolor”:”white”,”pinterest_enabled”:false,”caption”:null,”credit”:”SOPA Images\/LightRocket via Gett”,”focal_area”:{“top_left_x”:2100,”top_left_y”:1269,”bottom_right_x”:2900,”bottom_right_y”:2069},”bounds”:[0,0,5000,3337],”uploaded_size”:{“width”:5000,”height”:3337},”focal_point”:null,”image_id”:72127270,”alt_text”:””},”group_cover_image”:null,”picture_standard_lead_image”:{“ratio”:”*”,”original_url”:”https:\/\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg”,”network”:”unison”,”bgcolor”:”white”,”pinterest_enabled”:false,”caption”:null,”credit”:”SOPA Images\/LightRocket via Gett”,”focal_area”:{“top_left_x”:2100,”top_left_y”:1269,”bottom_right_x”:2900,”bottom_right_y”:2069},”bounds”:[0,0,5000,3337],”uploaded_size”:{“width”:5000,”height”:3337},”focal_point”:null,”image_id”:72127270,”alt_text”:””,”picture_element”:{“html”:{},”alt”:””,”default”:{“srcset”:”https:\/\/cdn.vox-cdn.com\/thumbor\/wV3tQxyxryDFrsomWmXRV0Ps-mg=\/0x0:5000×3337\/320×240\/filters:focal(2100×1269:2900×2069)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 320w, https:\/\/cdn.vox-cdn.com\/thumbor\/4NOA8AnqKURgbcJz9kBjA5SZBcw=\/0x0:5000×3337\/620×465\/filters:focal(2100×1269:2900×2069)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 620w, https:\/\/cdn.vox-cdn.com\/thumbor\/-NTZfdCm4hVfG_rglNASNDLUT08=\/0x0:5000×3337\/920×690\/filters:focal(2100×1269:2900×2069)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 920w, https:\/\/cdn.vox-cdn.com\/thumbor\/H6SSSfSetTEViabkg3L9EDSDXzk=\/0x0:5000×3337\/1220×915\/filters:focal(2100×1269:2900×2069)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 1220w, https:\/\/cdn.vox-cdn.com\/thumbor\/d4bUImjy3jfWML4aPYfaxgoeFkU=\/0x0:5000×3337\/1520×1140\/filters:focal(2100×1269:2900×2069)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 1520w”,”webp_srcset”:”https:\/\/cdn.vox-cdn.com\/thumbor\/lKOCFD6agdbVIQv_VmfKtxjEBCQ=\/0x0:5000×3337\/320×240\/filters:focal(2100×1269:2900×2069):format(webp)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 320w, https:\/\/cdn.vox-cdn.com\/thumbor\/Roo8ahsMSvpWc4O0YXXd-wpWD6A=\/0x0:5000×3337\/620×465\/filters:focal(2100×1269:2900×2069):format(webp)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 620w, https:\/\/cdn.vox-cdn.com\/thumbor\/ammjXN_v0ungYi4hDpZpkWEHkWE=\/0x0:5000×3337\/920×690\/filters:focal(2100×1269:2900×2069):format(webp)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 920w, https:\/\/cdn.vox-cdn.com\/thumbor\/HpSV7qwoSQI4-XbnZUS2dWexfZ0=\/0x0:5000×3337\/1220×915\/filters:focal(2100×1269:2900×2069):format(webp)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 1220w, https:\/\/cdn.vox-cdn.com\/thumbor\/F8ATNgKnw_N0hoHQPryvZ_Hub5A=\/0x0:5000×3337\/1520×1140\/filters:focal(2100×1269:2900×2069):format(webp)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg 1520w”,”media”:null,”sizes”:”(min-width: 809px) 485px, (min-width: 600px) 60vw, 100vw”,”fallback”:”https:\/\/cdn.vox-cdn.com\/thumbor\/O062Z5JY3SWZ2n5mN0uuY2GVWhk=\/0x0:5000×3337\/1200×900\/filters:focal(2100×1269:2900×2069)\/cdn.vox-cdn.com\/uploads\/chorus_image\/image\/72127270\/GettyImages_1249214710.0.jpg”},”art_directed”:[]}},”image_is_placeholder”:false,”image_is_hidden”:false,”network”:”vox”,”omits_labels”:true,”optimizable”:false,”promo_headline”:”How to test what an AI model can \u2014 and shouldn\u2019t \u2014 do”,”recommended_count”:0,”recs_enabled”:false,”slug”:”future-perfect\/2023\/3\/29\/23661633\/gpt-4-openai-alignment-research-center-open-philanthropy-ai-safety”,”dek”:”Inside the labs that helps evaluate AI safety for models like GPT-4″,”homepage_title”:”How to test what an AI model can \u2014 and shouldn\u2019t \u2014 do”,”homepage_description”:”Inside the labs that helps evaluate AI safety for models like GPT-4″,”show_homepage_description”:false,”title_display”:”How to test what an AI model can \u2014 and shouldn\u2019t \u2014 do”,”pull_quote”:null,”voxcreative”:false,”show_entry_time”:true,”show_dates”:true,”paywalled_content”:false,”paywalled_content_box_logo_url”:””,”paywalled_content_page_logo_url”:””,”paywalled_content_main_url”:””,”article_footer_body”:”Vox’s journalism is free because we believe that everyone deserves to understand the world that they live in. That kind of knowledge helps create better citizens, neighbors, friends, parents, consumers and stewards of this planet. In short, understanding benefits everyone. You can join in on this mission by making a financial gift to Vox today. Reader support helps keep our work free, for everyone. <http:\/\/vox.com\/pages\/support-now?itm_campaign=better-stewards&itm_medium=site&itm_source=article-footer>Will you join us? <\/a>“,”article_footer_header”:”<a href=\"http:\/\/vox.com\/pages\/support-now?itm_campaign=47-annual&itm_medium=site&itm_source=article-footer\">We have a request<\/a>“,”use_article_footer”:true,”article_footer_cta_annual_plans”:”{rn “default_plan”: 1,rn “plans”: [rn {rn “amount”: 95,rn “plan_id”: 74295rn },rn {rn “amount”: 120,rn “plan_id”: 81108rn },rn {rn “amount”: 250,rn “plan_id”: 77096rn },rn {rn “amount”: 350,rn “plan_id”: 92038rn }rn ]rn}”,”article_footer_cta_button_annual_copy”:”year”,”article_footer_cta_button_copy”:”Yes, I’ll give”,”article_footer_cta_button_monthly_copy”:”month”,”article_footer_cta_default_frequency”:”annual”,”article_footer_cta_monthly_plans”:”{rn “default_plan”: 1,rn “plans”: [rn {rn “amount”: 9,rn “plan_id”: 77780rn },rn {rn “amount”: 20,rn “plan_id”: 69279rn },rn {rn “amount”: 50,rn “plan_id”: 46947rn },rn {rn “amount”: 100,rn “plan_id”: 46782rn }rn ]rn}”,”article_footer_cta_once_plans”:”{rn “default_plan”: 0,rn “plans”: [rn {rn “amount”: 20,rn “plan_id”: 69278rn },rn {rn “amount”: 50,rn “plan_id”: 48880rn },rn {rn “amount”: 100,rn “plan_id”: 46607rn },rn {rn “amount”: 250,rn “plan_id”: 46946rn }rn ]rn}”,”use_article_footer_cta_read_counter”:true,”use_article_footer_cta”:true,”featured_placeable”:false,”video_placeable”:false,”disclaimer”:null,”volume_placement”:”lede”,”video_autoplay”:false,”youtube_url”:”http:\/\/bit.ly\/voxyoutube”,”facebook_video_url”:””,”play_in_modal”:true,”user_preferences_for_privacy_enabled”:false,”show_branded_logos”:true}”><\/p>\n<div>\n<p><strong><a href=\"http:\/\/vox.com\/pages\/support-now?itm_campaign=47-annual&itm_medium=site&itm_source=article-footer\">We have a request<\/a><\/strong><\/p>\n<p>\n Vox’s journalism is free because we believe that everyone deserves to understand the world that they live in. That kind of knowledge helps create better citizens, neighbors, friends, parents, consumers and stewards of this planet. In short, understanding benefits everyone. You can join in on this mission by making a financial gift to Vox today. Reader support helps keep our work free, for everyone. <http:>Will you join us?<br \/>\n <\/http:><\/p>\n<\/p><\/div>\n<div>\n<div>\n<p><label tabindex=\"0\" role=\"radio\" aria-checked=\"true\"><\/p>\n<p>\n <span>$95<\/span><span>\/year<\/span>\n <\/p>\n<p> <\/label><\/p>\n<p> <label tabindex=\"0\" role=\"radio\" aria-checked=\"true\"><\/p>\n<p>\n <span>$120<\/span><span>\/year<\/span>\n <\/p>\n<p> <\/label><\/p>\n<p> <label tabindex=\"0\" role=\"radio\" aria-checked=\"true\"><\/p>\n<p>\n <span>$250<\/span><span>\/year<\/span>\n <\/p>\n<p> <\/label><\/p>\n<p> <label tabindex=\"0\" role=\"radio\" aria-checked=\"true\"><\/p>\n<p>\n <span>$350<\/span><span>\/year<\/span>\n <\/p>\n<p> <\/label><\/p>\n<p> <label tabindex=\"0\"><\/p>\n<p> <span>Other<\/span><br \/>\n <\/label>\n <\/p>\n<\/p><\/div>\n<p> <a href=\"https:\/\/vox.memberful.com\/checkout?plan=\" id=\"contribute--submit\"><\/p>\n<p>\n Yes, I’ll give $120<span>\/year<\/span>\n <\/p>\n<p> <\/a><\/p>\n<p>\n Yes, I’ll give $120<span>\/year<\/span>\n <\/p>\n<div>\n<p>\n <span><br \/>\n We accept credit card, Apple Pay, and<br \/>\n <\/span><br \/>\n <span><br \/>\n Google Pay. You can also contribute via<br \/>\n <\/span>\n <\/p>\n<p><a href=\"https:\/\/www.paypal.com\/donate\/?hosted_button_id=VSP4PYJX98SHL\" target=\"_blank\" rel=\"noopener\"><br \/>\n <img decoding=\"async\" src=\"https:\/\/cdn.vox-cdn.com\/uploads\/chorus_asset\/file\/22734206\/paypal_logo.png\"><br \/>\n <\/a>\n <\/p>\n<\/div><\/div>\n<\/p><\/div>\n<\/div>\n<p><a href=\"https:\/\/www.vox.com\/future-perfect\/2023\/3\/29\/23661633\/gpt-4-openai-alignment-research-center-open-philanthropy-ai-safety\" class=\"button purchase\" rel=\"nofollow noopener\" target=\"_blank\">Read More<\/a><br \/>\n Johnathon Howe<\/p>\n","protected":false},"excerpt":{"rendered":"<p>About six months ago, I decided to make AI a bigger part of how I spend my time as a reporter. The world of AI is evolving very, very fast. New releases seemingly every week are changing what it means to be a programmer, an artist, a teacher, and, most definitely, a journalist. There\u2019s enormous<\/p>\n","protected":false},"author":1,"featured_media":623951,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[38,23337,46],"tags":[],"class_list":{"0":"post-623950","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-model","8":"category-shouldnt","9":"category-technology"},"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts\/623950","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/comments?post=623950"}],"version-history":[{"count":0,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts\/623950\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/media\/623951"}],"wp:attachment":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/media?parent=623950"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/categories?post=623950"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/tags?post=623950"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}