Replying to
@MsDropbear42@infosec.space
Agreed. I'd prefer them to work the word "simpered" into the description.
Remote
Donald Hobern
@dhobern@scicomm.xyz
mastodon 4.7.3Biodiversity informatician, naturalist/entomologist/lepidopterist, Java/Python programmer, Ngunnuwal Country / Canberra, AU
Australian Plant Phenomics Network Data Management Director
Species 2000/Catalogue of Life International Engagement Officer & Chair of Taxonomy Working Group
Working to automate insect monitoring solutions.
Was: ALA Director; GBIF Director; TDWG Chair; IBM software engineer
Banner: #PlumeMoth, Gillmeria ochrodactyla, Copenhagen, DK, 23 June 2014
He/him
0 Followers
0 Following
50 Posts
Joined November 06, 2022
Website:
iNaturalist:
Github:
Today I found six observations in #iNaturalist that include what may well be the only photos ever taken of live many-plume #moths in the genus #Triscaedecia.
This is a rarely recorded genus for which I accidentally became co-author on a 2019 global revision: https://www.researchgate.net/publication/332013222
There are six named species, four first described in the 2019 paper.
The new records include the second known individual of Triscaedecia svetlanae, extending its range from mainland #Malaysia to #Sabah in #Borneo, and the first (five) records of the genus from #Australia (far north #Queensland). The latter are probably Triscaedecia septemdactyla, a species previously recorded from New Guinea and the Bismarck Islands. I need to check the national #insect #collection here in #Canberra to see if there are any existing specimens in the unsorted material.
Replying to
@futurebird@sauropods.win
Australia has a system where you can be added to a self-exclusion list for gambling services hosted in AU.
Anecdotally, I hear that being on this list results in more spam from overseas online gambling services that are not covered by the exclusion.
Replying to
@MsDropbear42@infosec.space
And while we're at it, all factual assertions from politicians should be tagged with "opined the Minister [or whoever]".
Replying to
Replying to
@aj@gts.sadauskas.id.au @adavid@aus.social @spriebsch@phpc.social @preinheimer@phpc.social @pluralistic@mamot.fr
Probably the biggest opportunity for them will be selling propaganda and disinformation services to those who want to control society.
Replying to
@joannaholman@aus.social @futurebird@sauropods.win
The anecdotal example I gave was from someone who used that option. It works from a list of known gambling payment destinations, but knows nothing of the frequently ephemeral overseas examples. And it seems also that there are services that use PayID email addresses to route funds to selected gambling sites. It's good to have options like these, but they can give a false sense of security to concerned friends and relatives.
Replying to on tiny.tilde.website
@pho4cexa@tiny.tilde.website
"It still won’t be sustainable in the long run, but it doesn’t have to be: we just need to outlast the Crawlers. If you’ve read this blog, or follow me on the Fediverse, you’ll know there’s enough spite left in me to see us through."
That same spite is increasingly motivating a lot of my personal IT-related energy.
As I mentioned the other day, I really wanted an ultra-simple way to declare the contents of a slideshow presentation without getting sucked deep into Node.js hell.
So, I now have a #Python / #YAML / #markdown / #bash / #ImageMagick tool that does what I need and pretty much nothing more.
The code is in a @Codeberg@social.anoxinon.de repository: https://codeberg.org/dhobern/presentation
The YAML definition for my upcoming presentation is here: https://codeberg.org/dhobern/presentation/src/branch/main/slides.yaml
And here is the PDF generated from the YAML: https://codeberg.org/dhobern/presentation/src/branch/main/Lessons_from_Malaise_trapping_in_Aranda.pdf
Replying to
@indie@infosec.exchange
I've got no idea where you are but, if you're in the western US, these look like spotted fritillaries:
https://www.inaturalist.org/observations?preferred_place_id=6744&taxon_id=77171
Replying to
@jalefkowit@vmst.io
Grim, but I'm more interested in the decision to deploy autonomous lawn mowers and snow blowers inside private homes.
Replying to
@david_coombe_cbr@aus.social @RanaldClouston@fediscience.org
Thanks - books I'm interested in reading are rarely available within the ACT Libraries system - I've just reserved one of these.
I've got nothing positive to say about the colonialist programme with which he's most associated but #Kipling's short stories are a marvel. His prose is so controlled and modern, he throws us into minimally specified contexts, and he can sketch a character or mood with just a few words.
One example I've always loved is his Dayspring Mishandled (from Limits and Renewals). There's so much going on with this story.
Replying to
@nyhan@fediscience.org @anildash@me.dm
Any worthwhile model needs a way that its fit to some criterion of truth can be measured.
Many machine learning models are about classifying data or predicting values in ways that can be tested. Then the performance of the model against reality can be defined and documented.
With generative AI, there is rarely any objectively measurable criterion to assess the quality of the results other than the vagaries of user taste. This is what has led to so much over-hyping of mediocre result-shaped outputs.
Replying to
@joosteto@mamot.fr @miclgael@hachyderm.io
I'm now on Codeberg for all my own repositories (as dhobern), but they're all rather solo things, so hanging around with me would not be cool.
Replying to
@mcc@mastodon.social @mntmn@mastodon.social @cwebber@social.coop
I think there's a broader corollary (or perhaps it's actually a central subset of what you describe).
I always thought most people shared my experience that the exciting thing about the Internet, and good Internet search in particular, was that it offered access to the most relevant sources of information for any query. It was then on me to assess these sources and try to understand the topic at hand.
LLMs have resoundingly demonstrated that for most people this is all too much work and reminds them of school.
A majority of people clearly don't want to have to put in so much effort. They'd rather have an unambiguous answer that comes back and that they can treat as authoritative.
Sidenote - this is why mansplaining is a thing.
So, the primary (and I would argue, intended) result of the current "AI" mania is that the world is happily replumbing all its information and knowledge streams so that everyone receives whatever sanctioned propaganda those behind the curtain want to shovel out. (Pick a metaphor and stick with it ...)
LLMs are an assault on human communication and our ability to reason, organise and plan. They are the oligarch's wet dream.
Replying to
@sister_ratched
I hate those letters. So smug about nothing and completely ignoring everything I do care about.
Replying to
@baardhaveland @globalmuseum
I guess the question is whether natural nesting holes ever get cleaned out.
Replying to
@MostlyBlindGamer
I'm not sure it's particularly interesting or useful in itself.
I think my behaviour is affected by multiple things:
Expected number of items - if I know it will be in the range 1-3, I would emphasise the numbers from index to ring finger rather than from thumb.
Physical feedback - I think the reason I start with my little finger when truly counting is that I snap the fingers out from under my thumb which reinforces the sense of moving to a new state.
Social context - probably overrides all else.
Replying to
@MostlyBlindGamer
When I'm counting for my own benefit and the number is in the range 0-10, 1-5 are right-little to right-thumb, and 6-10 are left-thumb to left-little. I am right-handed.
If I'm emphasizing numbers while talking, 1-5 is right-thumb to right-little, and 6-10 is again left-thumb to left-little.
Replying to
70% of the way through the book and starting on CAN's The Lost Tapes. The opening track, Millionenspiel, is fantastic:
https://www.youtube.com/watch?v=u4CtYbTT2-Y
Can - Millionenspiel (from The Lost Tapes)
Replying to
@dhobern@biodiversity.social
The larger, furry black-and-white #moth in frames 3 and 4 is a female of the species Oenosandra boisduvalii (Boisduval's Autumn Moth).
It's the type species for the family #Oenosandridae, which is restricted to #Australia and includes only eight described #species.
This is a sexually dimorphic species, with the males mainly a sooty grey but with a surprising dusting of orange scales.
https://en.wikipedia.org/wiki/Oenosandra
https://www.catalogueoflife.org/data/taxon/BW9XR
#entomology #Lepidoptera #Canberra
Global #Lepidoptera Index updated. Feeds into #CatalogueOfLife & @gbif@biodiversity.social
https://www.checklistbank.org/dataset/55434/about
Complete update for #Apatelodidae and thousands of names updated in #Noctuidae
Names updated:
Noctuidae: 4787 Apatelodidae: 460 Erebidae: 338 Notodontidae: 25 Crambidae: 25 Pyralidae: 21 Autostichidae: 20 Pieridae: 20 Lycaenidae: 13 Riodinidae: 11 Nymphalidae: 7 Hepialidae: 7 Phiditiidae: 7 Zygaenidae: 7 Geometridae: 6 Nolidae: 6 Lasiocampidae: 5 Tortricidae: 4 Bombycidae: 4 Lypusidae: 3 Sphingidae: 2 Thyatiridae: 2 Uraniidae: 2 Yponomeutidae: 2 Limacodidae: 2 Oecophoridae: 2 Depressariidae: 1 Mimallonidae: 1 Hesperiidae: 1 Tineidae: 1
Major sources:
Ronkay et al. 2001, Hadeninae II
Hacker et al. 2022, Moths of Africa – Systematic and Illustrated Catalogue of the Heterocera of Africa, Volume 3 Acontiinae (Noctuidae) (Including Aediini Beck, 1960, Based on Traditional Treatment)
Ronkay et al. 2023, The Jacques Plante Noctuidae Collection. Part II. Amphipyrinae, Psaphidinae, Cuculliinae, Oncocnemidinae, Acontiinae, Pantheinae, Dyopsinae, Raphiinae, Acronictinae, Bryophilinae, Heliothinae, Condicinae & Xyleninae
Pohl et al. 2023, Annotated Taxonomic Checklist of the Lepidoptera of North America, North of Mexico
Orlandin et al. 2024, Systematics of Apatelodidae Neumoegen & Dyar, 1894 (Lepidoptera: Bombycoidea) based on molecular and morphological data, https://doi.org/10.1071/IS24041
Ronkay et al. 2017, Xyleninae I. The Agrochola generic complex
Volynkin et al. 2025, On the taxonomy of the genera Katha Moore, Tarika Moore, Cernyia Bucsek and Churingosia Volynkin & Černý, stat. n. with the description of a new genus and thirteen new species (Lepidoptera: Erebidae: Arctiinae: Lithosiini: Lithosiina), https://dx.doi.org/10.37828/em.2025.87.1
Fibiger et al. 2007, Amphipyrinae - Xyleninae
Hreblay et al. 1998, Noctuidae. In: Haruta, T. (Ed.), Moths of Nepal. Part 5., http://publ.moth.jp/special/Moths-of-Nepal-part-5.pdf
Volynkin et al. 2025, On the taxonomy of the Brunia Moore generic complex with the description of a new genus, a new subgenus and four new species (Lepidoptera: Erebidae: Arctiinae: Lithosiini: Lithosiina), https://dx.doi.org/10.37828/em.2025.87.2
Lafontaine et al. 2010, Review of the New World genera of the subfamily Acontiinae (Lepidoptera, Noctuidae), https://doi.org/10.3897/zookeys.39.427
Ronkay et al. 2011, Cuculliinae II and Psaphidinae
Hreblay et al. 1999, Neue trifide Noctuidae aus dem himalayanischen Raum und der südostasiatischen Region (Lepidoptera: Noctuidae)
Replying to
@ehasbrouck@urbanists.social @Yuvalne@433.world
Aren't you (and NWU there) describing the IA National Emergency Library (which can certainly be debated) rather than how IA has operated outside a ~3-month period in 2020?
Are you suggesting that IA is currently doing this?
Replying to
@ngaylinn@tech.lgbt @ftranschel@norden.social @DrewKadel@social.coop
Theatre is an excellent description. There's clearly NLP in front of the prediction engine itself, and some application of post-predictive review steps. But it's all presented as though this composite application is the LLM. It's like attending a seance.
Replying to
@ehasbrouck@urbanists.social @Yuvalne@433.world
Thanks again. The URL-per-page is a technical aspect (one that I rely on daily for managing references to materials in the #BiodiversityHeritageLibrary - part of IA, but separate from the OpenLibrary). The issues at question are the channels by which books become scanned copies, the proportions of such texts that can be accessed by users, and whether IA is operating properly as a library in what its doing for the still-encumbered works.
I do understand your concern and won't try to argue you into seeing IA as a positive, although I am confident it is one of the most important efforts we have for ensuring the preservation of human knowledge and culture in increasingly dark times. I do believe that the NWU page weakens its case by repeatedly insinuating bad faith (e.g. in the comments following the take-down of the freelance writers' guide).
Replying to
@threatresearch@infosec.exchange
In Australia - not seeing any such options under Data and Privacy.
The Permissions You've Given section just lists using Google to show last four characters of credit cards and something about making it easier to check out quickly. Neither seems to be something you can turn off.
Replying to
@davidgerard@circumstances.run
This is such a full summary of what's wrong with "AI" and expresses exactly how I feel about it.
Replying to
@ehasbrouck@urbanists.social @Yuvalne@433.world
Thank you - although that's the same link, and the unlimited-numbers-of-copies part in your first message relates only to a short period during the pandemic.
The part to which I can understand authors objecting (and even then IA is not the real cultural enemy) is the business about IA sharing scans for in-copyright books without paying the higher fee that libraries do for sharing copies.
Do you have any figures for what fraction of in-copyright books appear in IA? Pretty much everything I access is either out of copyright or else provided by the publishers/authors for public-good reasons. And the ones that I've seen loaned seem only to have minimal sample pages otherwise accessible (far fewer pages than e.g. a typical Amazon sample view). For example, is IA sharing copies of the books you've authored? Can you provide some concrete examples of books which illustrate the harm? Thanks.
Replying to
@tomw@mastodon.social There is a similar risk for every field that increasingly relies on citizen science data. Biodiversity research relies heavily on field observations to document the occurrence of species. This can have implications for conservation and land use and hence be of interest to extractive industries, etc. It seems inevitable that we will start seeing fake AI-generated observations that seek to obfuscate actual distributions of threatened species and communities.
Replying to
More subtly, could bad-faith actors compromise our systems by generating fake but plausible data to conceal actual patterns? AI chat tools seem perfect for such attacks. A corporation might try to prolong current operations and maintain profits by generating fake survey datasets or large volumes of bogus citizen science records that conceal habitat quality, minimise evidence of harm or fake the size of species populations. How would we detect and respond to such deception? (6/6)
Replying to
@Obdurodon@hachyderm.io
Wouldn't that just print "0", "fizzbuzz" 14 times, then "15", 14 more instances of "fizzbuzz", then "30", etc.? Isn't it missing three ! characters?
Replying to
@joncounts @KorimakoEcology I've been assuming that the AI would be able to learn enough to generate mostly plausible and habitat-appropriate but fake records. Generating plenty of these would make the observer seem experienced and reliable. This would make it possible to produce others that indicate new populations of rare species outside sites proposed for use by extractive industries. In the case of #iNaturalist, the RG requirements would also require them to fake new photos.
Replying to
@KorimakoEcology Maybe we need to gamify corporate interests. For example, would Nike be prepared to pay big money to establish a monitoring scheme through a large biodiverse region shaped like their swoosh, just so it appears that way as a hot spot on distribution maps?
Replying to
@dhobern@biodiversity.social
A night of Oenosandra boisduvalii males ...
https://en.wikipedia.org/wiki/Oenosandra
Replying to
@gvwilson
“When you have a technology so powerful it can make the Godfather but with cats, or write an entire University essay with only 30-35 errors, or…make Saving Private Ryan but with cats, there’s simply nothing that can hold it back.”
Replying to
@cthos
Can anyone explain to me how exactly a system prompt or any other context is expected to be privileged in an LLM execution?
Is the prompt just more context but with a weighting multiplier so it has a disproportionate effect on the token prediction?
Or is it somehow used as a metalinguistic wrapper around a quoted form of the user prompt? Surely it can't be as simple as "Play the part of a character with the role described by and respond in character to someone who has come to you and asked ''." But, if the LLM is just responding in any way at all to some concatenation of tokens from its context, there is no real semantic structure to what is input.
Or does the LLM first process the system prompt from natural language to something closer to a vector of control parameters so that it can transform the effective matrix of predictive weights for token sequences?
Regardless, I'm completely at a loss to understand why this seems to anyone like a properly tunable way to control token prediction.
Replying to