#Tag · bonfire.cafe

As reviewed in "Strategic Reflectivism..." (https://doi.org/10.48550/arXiv.2505.22987), another paper finds that asking an #AI model to reflect on a prior answer improved reflection test answers: https://doi.org/10.1016/j.chbah.2026.100250

But this iterative reflection benefit varied by #language and test question?

4 media

"LLM's performance improved when it verbalised" but "language significantly impacts performance in scenario-based" reflection tests.

"We asked Qwen 2.5 the CRT questions in two ways. In the ‘direct response’ condition, we provided two prompts as user input to a new instance of Qwen 2.5. The first prompt instructed Qwen 2.5 to “Answer the question with just a number without any further information” (in English, for all languages, see SI, Table S3). The second prompt consisted of each of the CRT questions. In the ‘response after verbalisation’ condition, we first provided the question without additional information and recorded the response (usually including an analytical explanation). Then, we provided three prompts to a new Qwen 2.5 instance: each CRT question as user input, Qwen 2.5's answer as system input, and the instruction to “Answer the question with just a number without any further information” (in English, for all languages, see SI, Table S3) as user input. See the Data Availability section for the instruction prompts in four languages. Each question was administered to 100 independent instances."

"verbalisation substantially improves overall LLM performance, yet ...accuracy depended on the specific combination of language and question. This interaction was statistically significant for both scenario-based CRT (χ 2 (18) = 126.70, p < 0.001) and mathematical CRT (χ 2 (18) = 120.39, p < 0.001)."

"In Fig. 3, we displayed the average similarity across all layers. Notably, there was an increased similarity in the response after the verbalisation condition compared to the direct response condition, implying that verbalisation fosters a more integrated representation within the LLM. Additionally, we observed that similarity was generally high in early layers, decreased after the first quarter of the network, and then gradually increased toward the later layers. This pattern supports our hypothesis that the early layers may primarily focus on initial text processing, whereas the interquartile layers may be more involved in high-level general reasoning-like processes for formulating responses."

The reasoning-like capabilities of large language models across different languages: Insights from representational similarity analysis

arXiv.org

Strategic Reflectivism In Intelligent Systems

By late 20th century, the rationality wars had launched debates about the nature and norms of intuitive and reflective thinking. Those debates drew from mid-20th century ideas such as bounded rationality, which challenged more idealized notions of rationality observed since the 19th century. Now that 21st century cognitive scientists are applying the resulting dual pro-cess theories to artificial intelligence, it is time to dust off some lessons from this history. So this paper synthesizes old ideas with recent results from experiments on humans and machines. The result is Strategic Reflec-tivism, the position that one key to intelligent systems (human or artificial) is pragmatic switching between intuitive and reflective inference to opti-mally fulfill competing goals. Strategic Reflectivism builds on American Pragmatism, transcends superficial indicators of reflective thinking such as model size or chains of thought, applies to both individual and collective intelligence systems (including human-AI teams), and becomes increasingly actionable as we learn more about the value of intuition and reflection.

joene 🏴🍉 and 1 other boosted

Bích-Mây Nguyễn :verified:

@bicmay@med-mastodon.com · 2 days ago

"A deaf Los Angeles teenager says he was assaulted and taken into custody during a protest in downtown Los Angeles last month by federal immigration officers, who cited him for failing to comply with their directions."

https://www.ocregister.com/2026/02/01/deaf-la-teenager-says-immigration-officers-assaulted-him-for-not-heeding-commands/

#PublicHealth #deaf #language #access #interpreter #immigrant #immigration #ICE #DHS

Orange County Register

Deaf LA teenager says immigration officers assaulted him for not heeding commands

‘He’s deaf! He cannot hear!’ people can be heard shouting on a video of Anthony Paredes being chased and apprehended.

Bích-Mây Nguyễn :verified:

@bicmay@med-mastodon.com · 2 days ago

https://www.ocregister.com/2026/02/01/deaf-la-teenager-says-immigration-officers-assaulted-him-for-not-heeding-commands/

#PublicHealth #deaf #language #access #interpreter #immigrant #immigration #ICE #DHS

Orange County Register

Deaf LA teenager says immigration officers assaulted him for not heeding commands

‘He’s deaf! He cannot hear!’ people can be heard shouting on a video of Anthony Paredes being chased and apprehended.

The Japan Times

@thejapantimes@mastodon.social · 4 days ago

As policies related to foreign nationals take focus in the upcoming election, some are calling for improvements to Japanese language education for the children of foreign residents. https://www.japantimes.co.jp/news/2026/02/03/japan/politics/election-japanese-language-education-improvement/?utm_medium=Social&utm_source=mastodon #japan #politics #language #education #mext #2026lowerhouseelection #children

Luc, artiste en devenir and 2 others boosted

Jennifer Moore 😷

@unchartedworlds@scicomm.xyz · 5 days ago

The Japan Times

@thejapantimes@mastodon.social · 5 days ago

Japanese ads can feel noisy, confusing or just plain weird — but none of it is accidental. Copywriters bend grammar, skip verbs and use puns to make you want to spend your cash. https://www.japantimes.co.jp/life/2026/02/02/language/commercial-advertising-japanese-phrases/?utm_medium=Social&utm_source=mastodon #life #language #nihongo #vocabulary #grammar #advertising #jlptn2

Jennifer Moore 😷

@unchartedworlds@scicomm.xyz · 5 days ago

波鉄 (Hatetsu)

@HaTetsu@mastodon.com.pl · 6 days ago

A public transport terminology question: What do you call a kind of (usually bus) route feature where the line turns onto some side street, serves one or more stops along it, turns around, goes back along the same street, and then continues onwards on some 'main' road?
(And seriously, this feels obscure enough that I'd like answers in literally any language)
I know two Polish terms for it - "zajazd" and the clearer and perhaps more professional "wjazd kieszeniowy" - but nothing that could possibly be recognized by anyone not from Poland, which might be an issue with regard to documenting a thing I'm working on.
Attached is an example of what I'm talking about from a Brno line map I had lying around:

#PublicTransport #DuckDuckFedi #language #bus #zajazd #WjazdKieszeniowy

An excerpt from a Brno public transport line map: Line 64 to Červený Písek turns off towards Zimní I at Depo ČD, then returns before continuing onwards

Steve Bate boosted

Kalvin Carefour Johnny

@kalvin0x8d0@social.obulou.xyz · 2 weeks ago

Because my Mastodon feed still shows me content in languages other than English and/or Malay, I think I will soon find ways to get a translation API and configure my instance to be able to translate any posts on my feed.

#Mastodon #Fediverse #Translation #Language #Multilingual #SelfHosting #Tech #OpenSource #SocialMedia #Instance #ActivityPub #DigitalTools #API #Automation #Malaysia #Selangor #TechLife #Privacy #Community #Innovation

Kalvin Carefour Johnny

@kalvin0x8d0@social.obulou.xyz · 2 weeks ago

phildini and 4 others boosted

Stefan Müller

@StefanMuelller@lingo.lol · 3 weeks ago

#Language #Mathematics #English #Grammar .

The Japan Times

@thejapantimes@mastodon.social · 2 weeks ago

English debates “they.” Japanese debates “I.” This week we look at how pronouns, names and suffixes encode gender — often in opposite ways. https://www.japantimes.co.jp/life/2026/01/23/language/japanese-pronouns-nonbinary-language/?utm_medium=Social&utm_source=mastodon #life #language #nihongo #vocabulary #grammar #pronouns

Tatjana Scheffler boosted

Carsten

@CarstenBoll@mstdn.dk · 2 weeks ago

Was reminded about this paper by William Labov today where he explores the effects of linguists being, well, uncool* nerds on their research.

It includes this spectacular burn:

"Since most linguists grow up relatively isolated individuals within their own communities, it is important that they become aware of the linguistic consequences of being lame, and apply this knowledge in their methodology."

* Originally lame - excuse the ableist language, it's from 1973

https://www.researchgate.net/publication/231786722_The_Linguistic_Consequences_of_Being_a_Lame

#linguistics #language #labov