Login
You're viewing the mstdn.social public feed.
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 8:11 AM

    Maria Ressa talks about how in an Eritrean language, AI changes ‘smallpox’ to ‘syphilis’, and sometimes confuses ‘medicine’ and ‘insecticide’, which obviously has public health implications.

    This type of work ‘AI performance and safety issues in other languages and its impact on people’ is 50% of the work I do.

    I’m glad Maria Ressa is involved in this work. She’s now talking about chatbots, teenagers, suicide.

    “Civil society has been saying this for years. You now have more evidence now for what you’ve been saying.”

    un.org/independent-internation

    #UN #Geneva #AI

    💬 6🔄 264⭐ 245

Replies

  • 💬 1🔄 2⭐ 13
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 9:19 AM

    Libya: we know that African countries and those participating the least in greenhouse emissions today.. will also not be benefiting from AI and contributing to the emissions from data centers

    Just as we would refuse for our natural resources to be exploited by others, we do the same with data

    #UN #Geneva #AI

    💬 1🔄 14⭐ 22
  • 💬 2🔄 15⭐ 22
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 9:27 AM

    Eswatini:

    We meet in Geneva at a defining moment in humanity. AI is no longer a distant innovation.

    For countries like Eswatini, AI governance must begin with ‘technology must serve the people’.

    AI must not narrow the world into a single identity or values or homogeneity. This is especially important for small developing states, who must not become mere consumers of technology made elsewhere, technology made without regard for their needs.

    Algorithms must be required to adhere to human rights, safety and public interest standards. AI systems must not be permitted to self regulate. Human oversight must remain at the center. Children must be protected from addictive design and manipulation by systems they cannot understand.

    Voluntary enforcement is not enough.

    #UN #Geneva #AI

    💬 2🔄 17⭐ 36
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 9:36 AM

    "An experiment is being run on our own societies without a plan and without our consent,"

    UN Secretary General António Guterres said while calling for a global ban on ‘killer robots’

    #UN #Geneva #AI

    💬 2🔄 31⭐ 43
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:16 AM

    Spain:

    We are speaking on defending human rights against techno fascism

    About whether our society should move forward or back

    You can count on Spain in moving forward in regulating AI

    #UN #Geneva #AI

    💬 1🔄 17⭐ 36
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:20 AM

    China:

    AI governance is the shared task of all countries. Chinese president calls for equitable AI governance. China has consistently supported the UN as the primary channel for global AI governance.

    Regardless of size, strength or social system, China has the AI action plan for cooperation and stands ready to share AI capacity building especially with the global south

    Open source cooperation. Open source AI is a shared asset. Chinese open source models like DeepSeek and Qwen have lowered the barrier.

    AI also brings unforeseen risks. China firmly upholds balancing development and safety. China stands ready to work with all parties on frameworks, standards and norms based on international standards.

    China calls on all parties to strengthen solidarity.

    #UN #Geneva #AI

    💬 2🔄 1⭐ 11
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:27 AM

    Switzerland:

    The UN should not only define the rules but also be the example.

    In the accomplishment of its mandate for those who they are serving: member states and populations around the world.

    Freedom online coalition, which Switzerland presided over this year, states that

    Governance of AI must be firmly enshrined in international law and human rights, and elaborated through human centered multi party, multilateral systems.

    We also highlight that AI can be used for surveillance and other abusive ways that can reduce our democracy and human rights.

    In Geneva we will host the next global summit on AI.

    #UN #Geneva #AI

    💬 1🔄 6⭐ 14
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:29 AM

    France:

    AI must reflect the linguistic and cultural diversity of the world, fight misinformation, respect the environment.

    Open source models to democratize access.

    Common governance to allow the freedom to allow sovereignty and human rights.

    For countries facing energy problems we need to make sure we don’t worsen the environmental situation.

    France is focused on renewable energy with Mistral and other efforts.

    We call upon all partners to support international governance. France is committed to share its experience here and to support these objectives.

    #UN #Geneva #AI

    💬 2🔄 2⭐ 18
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:32 AM

    Botswana:

    We recognize the capabilities of AI and want to harness it across key sectors. However, like many developing countries.

    We face challenges here. A major constraint is lack of adequate digital infra and skills in developing nations.

    We trust this dialogue will create the space for how best to support developing countries. It must serve as a platform for collective action that responds to the priorities of developing countries, to bridge the digital divide between nations.

    Ethical considerations to protect human rights and uphold international safeguards. We call upon practical measures from this dialogue to assist developing countries.

    #UN #Geneva #AI

    💬 1🔄 1⭐ 11
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:39 AM

    Kenya:

    AI must expand opportunity for all countries and must not deepen the divide.

    Inclusion, rights, risks: for many developing countries we feel the divide in compute, data and language.

    Many AI systems do not reflect African language or contexts. If this continues we will remain consumers of tech we do not shape while bearing most of its harms.

    We need to participate more meaningfully in setting standards. We need fair financing. Especially where it affects people’s rights.

    We now offer 22 000 govt services to millions of citizens. Interoperability is the common baseline for safety.

    #UN #Geneva #AI

    💬 1🔄 2⭐ 12
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:42 AM

    Guatemala:

    We welcome this opportunity to ensure the tech transformation of all people instead of depending existing divides.

    We consider that ethical principles, human rights, scientific principles, international standards.

    Discussion is not enough. We must speak of connectivity, infrastructure, literacy and institutional strengthening for those countries facing structural difficulties.

    Our reality: we are plural multicultural and multilingual. A truly inclusive AI cannot be limited to expand technologies designed to work without cultural and linguistic diversity. Incorporate indigenous languages, incorporate community and ancestral knowledge, reduce bias, so as to not make our contributions invisible.

    Digital inclusion must also be cultural, linguistic, terroritorial inclusion. We must actively participate in defining standards, frameworks.

    #UN #Geneva #AI

    💬 1🔄 2⭐ 11
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:45 AM

    Lebanon:

    A long time ago our ancestors looked to the sea, used it as a bridge and not an obstacle. We exchanged cultures and goods.

    With the same spirit we look with some concern to the future of AI.

    We have known wars, displacement and loss, destroyed institutions.

    Our story is always about renewal and rejuvenation.

    We are investing in digital capacity building programs. We have revised the school curriculum to include responsible AI. We are working with Lebanese expatriates abroad to help build institutions, promoting homegrown innovation not in isolation but with choice and international cooperation.

    We need standards, safeguards. Lebanon is ready for a purposeful dialogue, to navigate a future with AI.

    #UN #Geneva #AI

    💬 1🔄 2⭐ 12
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:49 AM

    Tanzania:

    The central question we need to address is how we ensure inclusive and interoperable and beneficial digital transformation.

    Developing privacy and security so we can protect our citizens.

    We have national guidelines and policy frameworks.

    We work closely with ITU African Union and other partners to ensure systems can support our development goals while respecting privacy and security and sovereignty.

    Developing countries must have an equal voice. We call for enhanced international cooperation, research partnerships, investments in digital infrastructure so all countries can participate meaningfully.

    #UN #Geneva #AI

    💬 1🔄 3⭐ 11
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 10:58 AM

    Comments:

    Luisa Machado from Equilabs: can anyone younger than 30 years old stand up?

    I feel like we are 10-15 people here.

    Discussion about children and young people safeguards don’t represent us.

    Many young people are not excited about AI. We understand the risks.

    You talk about age verification and representing youth.

    You just want to govern young people’s relationship with technology instead of governing technology companies’ relationships with power.

    You continue to sign multi year contracts with companies that you are supposed to regulate.

    You have to move beyond the governing the symptoms.

    #UN #Geneva #AI

    💬 2🔄 54⭐ 75
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 11:02 AM

    Someone giving comments on language technology.. who has an accent.. had his ‘global alliance’ comment transcribed as

    ‘Global aliens’

    Yep

    #UN #Geneva #AI

    💬 3🔄 9⭐ 26
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 11:06 AM

    Comments from someone who works on understanding impact of AI on human cognition:

    We’ve talked a lot about technical and human governance but we need more urgency on how AI impacts how people think, as they over-rely on AI and demonstrate declining cognitive ability

    #UN #Geneva #AI

    💬 1🔄 8⭐ 14
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 11:07 AM

    Comments: what happens to low income countries when funded projects are over? Who owns the model and the data? Who ensures the work can continue even after the foreign funding dries out?

    This is especially relevant in philanthropic-funded healthcare projects in the global south.

    #UN #Geneva #AI

    💬 2🔄 14⭐ 14
  • Adrianna Tanskinnylatte@hachyderm.io
    Jul 6, 2026, 11:28 AM

    Barbados:

    Critical data has to stay within the country under our jurisdiction on infrastructure that we control.

    Countries should not have to pick between capability and independence / sovereignty.

    We ensure that a in-country team of trained data specialists that can own critical functions. We do this for sustainability.

    We believe in bringing communities at the design stage, not bringing models and expecting them to work. We believe in local context and co-design.

    #UN #Geneva #AI

    💬 0🔄 2⭐ 5
  • keeps vis.social runningkristinHenry@vis.social
    Jul 6, 2026, 10:05 AM

    @skinnylatte and no IRB! Ask anyone who does academic research that involves humans (and more).

    You can't even run experiments on color blindness with human subjects in academia without an IRB.

    💬 0🔄 2⭐ 3
  • 💬 0🔄 0⭐ 0
  • Alanmetaphase@toot.community
    Jul 6, 2026, 11:25 AM

    @skinnylatte I think for AI, an extension of the ideas around Protected Designation of Origin would be interesting (similar to how Champagne has some protection as a french origin term, Parmigiano Reggiano has an Italian DOC/PDO). In the AI form, Worker cooperatives built around AI training specialties and regulated IP regarding national, cultural, and topic areas would be interesting - e.g. is this the company maintaining an AI trained for and certified for French medical consultations?

    💬 1🔄 0⭐ 0
  • Alanmetaphase@toot.community
    Jul 6, 2026, 11:29 AM

    @skinnylatte Nations that are serious about these AI independence should provide legal worker coop organizational/corporate, IP, and financial framing to protect sovereignty by "enfranchising" the people and data into organizations for producing and maintaining operational AI technologies.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • slotosslotos@toot.community
    Jul 6, 2026, 12:18 PM

    @skinnylatte I really dislike misuse of “open source” when it comes to AI models.

    Source for those models is the entire corpus of training data with a full process of tuning them.

    What is peddled as open source is basically a binary executable that “you can read and tweak bytes of”.

    💬 0🔄 0⭐ 0
  • Urzlgooba42@mastodon.social
    Jul 6, 2026, 4:35 PM

    @skinnylatte For all its flaws, weirdly including suppressing knowledge of said flaws, it's really irritating how often China looks super reasonable in the era of Trump.

    "Sure, we murder and torture dissidents but at least AI-driven unemployment doesn't mean death here!" - This argument shouldn't work as well as it does.

    💬 0🔄 0⭐ 1
  • Daniel Silverstonekinnison@flarn.net
    Jul 6, 2026, 9:58 AM

    @skinnylatte One thing I don't get is why "Children must be protected from addictive design and manipulation by systems they cannot understand. " always invokes Children rather than **people**.

    Plenty of adults (I'd wager close to 100%) don't understand the manipulative tactics in play either.

    💬 2🔄 8⭐ 11
  • Michael Westergaardmichael@westergaard.social
    Jul 6, 2026, 10:35 AM
    The argument is that adults can make their decisions for themselves. Alcohol, tobacco, and sweet/fatty foods are also bad for you, yet legal.

    To argue that SoMe should be banned altogether you would have to argue it is closer to meth than a bottle of whisky. You cannot go around outlawing things just because you dislike them, or you better have a VERY good reason.
    💬 2🔄 0⭐ 0
  • Daniel Silverstonekinnison@flarn.net
    Jul 6, 2026, 11:01 AM

    @michael Right but the term "informed decision" requires that the adult be informed. Almost all of the deliberately addictive nature of SoMe is also deliberately obfuscated in order to reduce the chance that adults are properly informed.

    Make SoMe carry warnings like cigarette packets have to. No need to ban it, just make it unpalatable and unprofitable for these companies to exploit, rather than provide a reasonable service to, their users.

    💬 2🔄 0⭐ 2
  • Daniel Silverstonekinnison@flarn.net
    Jul 6, 2026, 11:03 AM

    @michael Things like requiring explicit enthusiastic ongoing consent for targeted advertising, or for an "algorithm" which is anything other than a chronological feed of everything the people who you directly follow have said/boosted would be a good start.

    "waah that would mean we'd lose advertising revenue" fine - offer a *fair* price to opt out of the advertising and "algorithm" all together.

    💬 1🔄 0⭐ 0
  • Daniel Silverstonekinnison@flarn.net
    Jul 6, 2026, 11:04 AM

    @michael If you can't run a business profitably without underhandedly exploiting your users, perhaps you shouldn't be running that business?

    💬 0🔄 0⭐ 0
  • Michael Westergaardmichael@westergaard.social
    Jul 6, 2026, 11:32 AM
    That I'm perfect ly fine with – to a degree. Such legislation would be careful to not make it like tobacco lung-pictures, which after an initial effect are largely ignored by smokers, or cookie-consent, which got perverted by advertisers from "don't gather all the data" to "annoy users with banners they won't read and then gather all the data."

    And it doesn't preempt the discussion about how to protect kids, because there's a good argument they cannot give informed consent.
    💬 1🔄 0⭐ 0
  • Daniel Silverstonekinnison@flarn.net
    Jul 6, 2026, 11:40 AM

    @michael I understand the point about children being unable to consent and appreciate they need special consideration as a result, and perhaps more immediate action; however *not* framing this as affecting everyone is unhelpful since it tacitly leaves SoMe et al. with permission and encouragement to further exploit unsuspecting adults instead. Eg. with allegedly "child safe" versions of the indoctrination machines.

    💬 1🔄 0⭐ 1
  • Michael Westergaardmichael@westergaard.social
    Jul 6, 2026, 12:04 PM
    I have a feeling big tech will use any argument to broaden the legislation to include everybody to further stall.

    At the same time, while I'm all for not blaming victims, people should also know that SoMe is bad for you by now – heck, most spend a lot of time complaining about doom-scrolling and hell-sites. It's almost as prevalent as knowing smoking is bad. At some point we have to accept people doing dumb things to themselves and that no amount of labeling will help.

    That doesn't mean people shouldn't have tools and information necessary to stop themselves, just like you can put yourself on a gambling blacklist in many places.
    💬 0🔄 0⭐ 0
  • Leelooleeloo@c.im
    Jul 6, 2026, 1:39 PM

    @michael @skinnylatte @kinnison
    Nobody argues for banning social media.

    The argument is that maybe the Epstein class shouldn't be the ones owning social media, and as long as they do, governments need to restrict how much intentional evil they are allowed to do.

    Rather than basically forcing people to be on Facebook and Xitter by making that the only way to participate in democracy, then banning the next generation of voters.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Advanced Persistent Teapothttp_error_418@hachyderm.io
    Jul 6, 2026, 9:33 AM

    @skinnylatte why do they have to lead with "The risk of humans losing control of AI systems is rapidly growing, the report said." when they immediately follow to say that it is actually humans controlling it, just for private ends instead of for social benefit 😔 does that irk you too? It feels like it's undermining the message of the report and reinforcing the people who suggest it has its own motive force and consciousness.

    💬 1🔄 0⭐ 3
  • Advanced Persistent Teapothttp_error_418@hachyderm.io
    Jul 6, 2026, 9:37 AM

    @skinnylatte huge congratulations for your achievement in getting to help deliver something like this, it sounds like the report itself is excellent. Just their PR copy about it that is a little dubious 😅

    💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 1
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 1
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Rr@glauca.space
    Jul 6, 2026, 12:44 PM

    @skinnylatte i can still easily break machine translation using written colloquial cantonese, and this language has 80+ million speakers

    💬 1🔄 0⭐ 1
  • Rr@glauca.space
    Jul 6, 2026, 12:50 PM

    @skinnylatte oh, and a bunch of the (free) chatbots still seem to struggle with "explain the correct stroke order for writing characters", even though we have machine-readable data files for this, that we've *been using* for "AI" (hidden markov models for handwriting recognition) on mass-market consumer computing devices since the 90s/00s

    💬 0🔄 0⭐ 1
  • Jesse McClure :progress_pride:jmcclure@sciences.social
    Jul 6, 2026, 1:12 PM

    @skinnylatte

    This is an intriguing issue I hadn't previously considered.

    I can see how an LLM could be much worse in a language with less training data, but do you know of any assessment of whether translating queries to English for the LLM then translating responses back to Eritrean would work at all?

    Not that it'd be a solution. The solution is just to not trust the damn bots in ths first place regardless of language, of course.

    💬 0🔄 0⭐ 0
  • draeathdraeath@infosec.exchange
    Jul 6, 2026, 3:37 PM

    @skinnylatte interesting, my wife (Turkish) calls insecticide "medicine" as well.

    I was really concerned when she was asking for me to find the "ant medicine" the other week!

    💬 0🔄 0⭐ 0