Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality
Anya is LIVE right now
FREE
Free to watch • No registration required • HD streaming
Text of tweet under the cut because it is loooong.
But... Stochastic Parrots.
Timnit Gebru was fired from Google in December 2020 for refusing to retract a research paper, and every single warning that paper made about large language models has now happened at a scale the industry spent 4 years trying to make people forget about.
Her name is Timnit Gebru.
She co-led the Ethical AI team at Google. She co-wrote a paper called "On the Dangers of Stochastic Parrots" with Emily Bender at the University of Washington and two other researchers. The paper was 14 pages long. It was submitted to a top AI ethics conference. And it was the reason Google decided that one of the most senior Black women in AI research could no longer work there.
The story Google told publicly was that she resigned. The story she told, confirmed by 2,695 of her colleagues in an open letter, was that she was fired by email while on vacation because she refused to either retract the paper or remove her name from it.
The paper had not even been published yet.
Here is what she actually wrote, and why every prediction inside it has now come true.
The first warning was about scale itself. Bender and Gebru argued that training ever-larger models on ever-larger scrapes of the internet would produce systems that appeared fluent but had no actual understanding of language. They called these systems stochastic parrots because they would repeat patterns from training data with statistical confidence and zero comprehension. The paper predicted that this apparent intelligence would fool both users and developers into trusting outputs that were structurally incapable of being reliable.
This was 2020. GPT-3 had just come out. The paper predicted the hallucination problem before anyone had a word for it.
The second warning was about bias amplification. The paper documented in detail that internet-scale training data contains systematic overrepresentation of dominant viewpoints and underrepresentation of marginalized ones. The models would not just absorb this bias. They would amplify it, because the optimization process rewards confident outputs, and confidence in language patterns tracks frequency in the training set.
The prediction was that hiring tools built on these models would discriminate against women. That healthcare triage tools would underperform on Black patients. That loan approval systems would entrench inequality while presenting their decisions as neutral algorithmic judgment.
Every one of those things has now been documented in deployment.
Amazon's hiring algorithm penalized resumes that contained the word "women" in any context. Healthcare risk scoring algorithms used by major US hospitals were found to systematically underestimate the medical needs of Black patients. Apple Card's credit algorithm gave wives credit lines 10x lower than their husbands for the same financial profile.
The third warning was about environmental cost. The paper calculated that training a single large language model produced emissions equivalent to the lifetime output of 5 cars. The prediction was that the race to scale would create an environmental footprint that would eventually rival entire industries.
In 2024, Google's emissions were up 48% from 2019, and the company explicitly blamed AI infrastructure. Microsoft's were up 29%, same reason. Both companies have now quietly abandoned the climate commitments they were publicly celebrating the year Gebru was fired.
The fourth warning was about documentation. The paper argued that the training datasets being assembled were too large for anyone to actually audit. Nobody at Google, OpenAI, Meta, or any other lab could tell you with confidence what was in the data their models were trained on. This was not a temporary problem to be solved later. It was a permanent feature of the approach.
In 2023, researchers discovered that the LAION-5B dataset, used to train Stable Diffusion and other major image models, contained thousands of images of child sexual abuse material. The companies that had trained on the dataset had no way of knowing. The paper predicted that category of failure 3 years before it was found.
The fifth warning was the one Google cared about most.
Bender and Gebru argued that the deployment of these systems would centralize linguistic and cultural power in the hands of the small number of companies that could afford to train them. The internet would become a place where the dominant voice was a statistical average of dominant voices, presented as a neutral assistant. Languages underrepresented in the training data would degrade over time as more web content was generated by these systems and fed back into the next training run.
This is now happening in real time. A 2024 study found that 57% of new web content in English is AI-generated or AI-assisted. Researchers studying low-resource languages have documented active degradation in translation quality, because the synthetic content fed back into training is itself worse in those languages.
The paper Google fired her for predicted the model collapse problem before model collapse had a name.
The mechanism behind why this all happened is the part of her work that nobody quotes.
Gebru's argument was not that AI is dangerous in some abstract sci-fi sense. Her argument was that AI is dangerous in a very specific structural sense. The technology was being built by a small group of researchers who shared similar backgrounds, worked at similar companies, and were rewarded for shipping products faster than competitors. The incentive structure made it impossible for safety, ethics, and bias concerns to slow anything down. Anyone inside the system who raised those concerns was either ignored, sidelined, or removed.
She was making that argument from inside Google.
Then Google proved her right by removing her.
The team Google had built to make sure their AI was safe was dismantled in 90 days because they did the job they had been hired to do. Margaret Mitchell, the other co-lead of the Ethical AI team, was fired two months after Gebru for searching through her own emails for evidence of how Gebru had been treated.
Gebru did not stop. She founded DAIR, the Distributed AI Research Institute, in 2021. The mission is to do AI research outside the control of the companies that have a financial interest in not hearing the answers.
Every prediction in the Stochastic Parrots paper has now been validated by deployment. Hallucinations are an industry-wide problem the largest labs cannot solve. Bias amplification has been documented in hiring, healthcare, lending, and criminal justice. Environmental costs are larger than entire small countries. Training data audits remain impossible. Model collapse is an active research crisis at every major lab.
The question worth sitting with is the one almost no one in the industry will say out loud.
Every researcher with the technical credibility to call out these problems watched what happened to her in December 2020 and made a calculation about their own career. The number of people willing to speak publicly about safety and ethics issues inside the major AI labs collapsed after that firing and has not recovered.
The researcher Google fired for warning about exactly what is now happening was right.
The company that fired her is now the second-largest deployer of the technology she warned about.
And the people inside that company who agree with her are not allowed to say so.
Reblogging for the purpose of reminding you all that you should not trust AI summaries of science. They can often be convenient! They can often be correct! But expecting perfection from a model fed by the imperfections of ongoing scientific work is a failure of the reader.
It's all about the superior colliculus (SC), a brain region involved in communicating the spatial location of significant objects based on sensory input. So it gets a bunch of sensory info and transforms it into motor plan suggestions-- if you're hungry, your SC can tell motor areas that there is a delicious, Scooby-Do-like sandwich sitting on the table a foot to your left. Before you know it, you're reaching for that sandwich and scarfing it down like a starved man. Probably.
This paper specifically looked at what the SC is up to during sugar self-administration in mice. Self-administration is important because it can be an animal model of addiction behaviors: animals will self-administrate more of an addictive substance than a non-addictive substance. The authors found that neurons in the SC are more active when re-exposed to a visual signal (in this case, a light) that was shown during sugar self-administration. The brain has learned that light = sugar, so when they see the light, the SC goes AHA!! An important thing!!
They also found that inactivating these neurons prevented the animal from self-administrating as much sugar, suggesting that without the visual signal telling them that they were being rewarded, they were less likely to go for the reward.
All of this can be used to further understand how the SC might play a role in addiction by flagging the cues associated with rewarding behavior-- like how you're far more likely to relapse into drug use if you can see drug paraphernalia.
The details of the methods in this paper were a little hard to understand, I'll be honest, but I think that was a language barrier issue. The authors are all from Beijing or Nanjiang in China, and I am unable to read Mandarin, so we had to settle for an English version.
See you next time!
Cun, X., Zhu, H., Wu, N., Jing, M., & Song, R. (2025). Superior colliculus encodes the visual cue associated with rewarding behavior in sucrose self-administration in mice. Scientific Reports 2025(15):35171. https://doi.org/10.1038/s41598-025-19192-3
Ohoho today's paper is not the most revolutionary, but it IS a great example of one of my favorite forms of science communication: following the questions!! The paper is called "The what, which, when, why and who of Off responses in the auditory system," and boy howdy does it answer What? Which? When? Why? and Who?
What makes these happen: "Off" responses are neuronal signals that specifically occur after a stimulus has disappeared. In the auditory system, these are signals that start blaring once the sound ends. A lot of research focuses on signals from when the sound starts (On responses), so these Off responses are under-studied. This section discusses What is causing the signals, and it can't come to a definitive conclusion because there's a lot of slightly contradictory results out there, but it seems that Off responses are present in most pit stops on the auditory processing pathway. By the time you get to the cortex, there are probably multiple kinds of inputs from upstream areas making their Off responses complicated.
Which stimuli affect them: Off responses seem to be sensitive to the spectrotemporal properties of the sound, which is a fancy way of saying How Complicated the Sound Is and How It Changes Over Time. It makes sense that an Off response would do this, because On responses literally haven't had time to do these fancy calculations as a sound changes. They have to start immediately.
When do Off responses affect behavior: uhh whenever an animal needs to know that a sound stopped ig
Why do we need them: since Off responses have access to sound information over time (compared to On responses), they're great for checking if a sound was behaviorally relevant!
Who in the brain is making them: it seems rn that very brief, very fast Off responses are probably coming from early in the pathway, while longer, later responses are probably coming from later in the pathway. Generally speaking, the later the area in a pathway, the more information it's working with-- so perhaps the fast Off responses are literally just "sound's gone!" while the longer ones are more like "the sound ended, and it seems to be indicating that my offspring needs food right now." Since cognitive stuff can get tied in there.
The reason I like this style of paper so much is because it's such a genuine representation of how scientific thinking works. You approach a concept-- Off responses-- and then explore based on what questions you have. What's happening with Off responses? What are they responding to? When do we use them? Why do we use them? Where are they coming from? Structuring a paper as "we did this, which prompted the next question" feels organic and logical in a way that some paper structures don't. It is much easier to follow complicated results when they read like a narrative.
Anyway, good paper! It's a review article, so potentially somewhat accessible since you don't have to get bogged down in the experimental details.
Ciao!
Edeline, J.-M., & Liu, R.C. (2025). The what, which, when, why and who of Off responses in the auditory system. J Physiol 0.0:1–12. DOI: 10.1113/JP289100
Today's paper summary is something thrilling: incomplete. This is because I did a surgery in lab this morning and now I'm completely exhausted and the words are swimming on the page.
I read most of it though, and I had a difficult time understanding it. Upon deciding to give up, I realized why this may have been: the authors appeared to be mainly from engineering departments. Now I have nothing against engineers, but it can be obvious when a paper from your own field has a sort of accent to it, and this one was totally an engineer's take on neuroscience. And this is super valuable! Seeing how other people solve problems helps you solve your own more creatively. But also I am too tired to figure out what your graphs are doing.
The paper is about neurons in the auditory cortex, and their responses to high and low contrast sounds in an oddball task (a bunch of standard stimuli, then bam, new one!). They were also pretty focused on inhibitory neurons, specifically ones that expressed the proteins parvalbumin and somatostatin. So although my summary here is incomplete, at least you have learned about parvalbumin and somatostatin as markers of inhibitory neurons.
Give this one a read if you have the spine for it! I, a known coward, will be back later with something more my speed.
Ciao!
Mukesh, A., Mehra, M., & Bandyopadhyay, S. (2025). Spectral Contrast and Context Preference in the Auditory Cortex Is Shaped by Specific Inhibitory Neuron-Based Subnetworks. European Journal of Neuroscience, 62:e70307. https://doi.org/10.1111/ejn.70307
Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality
Anya is LIVE right now
FREE
Free to watch • No registration required • HD streaming
Today's paper honestly wasn't that cool to me. Kinda cool, but not super cool.
They can't all be winners.
This paper is all about the superior colliculus, a brain region that helps combine sensory information into a useful map for motor systems to use when planning behavior. In a lot of animals, it's mainly a visual area with some auditory stuff, and it is hugely involved in planning saccades (little eye movements, lookin' around at stuff). This study was done in primates.
The study used a very delicately balanced task to see what part of the saccade process was most obviously represented in the superior colliculus. They found that accuracy (looking at the right thing) was not really the superior colliculus's job-- it was equally excited for right and wrong answers. However, the superior colliculus neurons got SUPER excited when something task-relevant appeared in their little "field of view" (these neurons don't "see," so this is called their "responsive field" or RF) regardless of whether it would lead them to a correct answer. What fun! They're little detectors that go HI HELLO THERE IS SOMETHING TO LOOK AT OVER HERE. CONSIDER LOOKING THIS DIRECTION. and presumably other regions down the line make the decision about whether looking at that thing would be a smart move or not. There's also visual, motor, AND visuo-motor neurons in the superior colliculus, with some differences in how the different subcategories respond to stuff. The big responses in this paper were from visuo-motor neurons.
Fancy stuff! I have nothing against this paper, I just found the task hard to follow and I'm distracted today (I have just barely averted a lab crisis and I'm still coming down from that panic).
Let me know what you think!
Ghosh, S., & Maunsell, J.H.R. (2026). Attention-related modulation in the superior colliculus encodes perceptual sensitivity, but not perceptual choice. Nature Communications, 2026(17):3323. https://doi.org/10.1038/s41467-026-69954-4
Today's paper: ...also auditory. But also visual. But the difference between the two groups is that one group is deaf.
Backing up, we have a lot of evidence that long-term sensory deprivation prompts remodeling in the brain. If you're not using your vision, then all those visual areas can be co-opted for something else. The brain is an efficiency machine. In this case, the paper was looking at how congenitally deaf people's auditory cortex is involved in visual processing.
They used fMRI (functional magnetic resonance imaging) for this study, which is an absolutely amazing but exceedingly tricky piece of technology. The general gist of it is that it uses MRI tech to detect changes in the blood oxygen level dependent (BOLD) response in the brain, with the assumption that areas using more oxygen are areas that are more activated during whatever is going on. Someone in an fMRI sees something = activates visual cortex = visual cortex needs more oxygenated blood = scanner detects BOLD response. Ta da!
What's interesting here, though, is that they found that a major difference between deaf and hearing participants was deactivation, or an area using less oxygenated blood. Specifically, they looked at the auditory cortex while participants were looking at visual stimuli. And bam, they found that deaf participants specifically had more deactivation of the auditory cortex!!
This could mean a lot of things. It does NOT, however, mean that deaf people aren't using their auditory cortex during vision. Deactivation doesn't mean an area isn't involved at all-- both positive and negative signals mean that something is changing in that brain area. But with so much emphasis put on activation, it's really cool to see an example of deactivation playing a big role!
Before reading this, I was not super aware of how negative BOLD responses could be useful for fMRI, so this was really neat!!
Wishing you all the best, as always, and treat your brains kindly today.
Ciao!
Tal, Z., Sayal, J., Fang, F., Bi, Y., Almeida, J., & Fracasso, A. (2026). The Neural Organization of Visual Information in the Auditory Cortex of the Congenitally Deaf. Human Brain Mapping, 2026; 47:e70444. https://doi.org/10.1002/hbm.70444
If you guys have had a couple of specific life experiences, like having a baby that requires a full diagnostic hearing test or maybe you personally requiring a full diagnostic hearing test, you may have come across the concept of an ABR. "ABR" stands for Auditory Brainstem Response, and it's all about the earliest levels of sound processing in the brain. Sound goes in the ears, where the air pressure wave information becomes electrical wave information via the cochlea, and then it goes up the brainstem while ping-ponging around a bit, through the inferior colliculus and thalamus (there's no quiz, don't worry), and to the auditory cortex! Hooray! The ABR specifically looks for trademark neural responses in the brainstem to ensure that sound information is become electrical wave information properly.
But don't get too excited, because the ABR can only check the first couple of stops on this Hearing Train. The cortex is super important for processing more complicated sounds, as well as making sure you're actually getting the useful info from the sounds, so the ABR alone can't tell the whole story. That's where today's paper comes in: attempting to change the parameters of the ABR test in bats to capture the slower responses from the auditory cortex, called the cortical auditory evoked potentials (CAEPs).
In short, they had five questions, which they promptly answered by messing around with ABR parameters:
If we measure longer, can we see evidence of cortical stuff AFTER the brainstem stuff? Answer: yup! The five-ish, six-ish waves of the ABR transition nicely into a neat positive peak (P1), negative peak (N1), and another positive peak (P2) after you play a sound to a bat. Hooray for the CAEP!
Do these cool new measurements respond differently when you mess with the sound? Answer: yup! Louder sounds make CAEPs bigger, while higher pitched sounds make them smaller (the lowest pitch they played was one that bats use a lot for echolocation, so it makes sense that higher pitches aren't as interesting to them).
If we play sounds really fast, will the CAEP get smaller? They're talking about a concept called "forward suppression," where one sound can diminish the response to a later sound. And sure enough, the faster they played these sounds, the smaller the cortical peak activity! They started playing sounds twice per second, then the CAEP got significantly smaller as they hiked it up to 8-16 times per second. These bats make echolocation calls at around 180 times per second at their fastest, so it's crazy to think about what the CAEP might be doing then.
If we change the types of sound, will the CAEP look different? Answer: yes! Unsurprisingly, the CAEP showed the biggest responses to downward frequency modulations (quick sweeps from high to low pitch, like a slide whistle-- this is the same style as their echolocation calls) and noise bursts (just a big BLAM of a bunch of different frequencies, like a super loud white noise machine). These are much more similar to natural sounds than the upward frequency modulations (a slide whistle going the other way) or pure tones (a single frequency-- this does NOT happen naturally).
What if we make it sound even MORE like echolocation, with pulse-echo pairs? Here they're pretending the bats are echolocating by playing both a vocalization sound AND the "echo" afterwards. Basically what they saw is that the initial ABR would happen twice, once for the vocalization and once for the echo, but the CAEP would just happen once even stronger than usual. They suggest that this shows how the auditory cortex is kind of adding together the really fast ABR responses, but this isn't perfectly linear, and the pattern doesn't hold for 5 echolocation sounds back to back. So it's not purely additive.
Overall, I think this paper slapped! It was (to me) straightforward and easy to read, and some cool results came out of it! The ABR and CAEP are both already used in humans, but they could be useful for bat research to help us tie together all these complicated auditory processes in an animal model.
Hope you guys enjoyed, and happy summer!!
Fouhy, V., Ellis, S., & Smotherman, M. (2025). Subcutaneous cortical auditory evoked potentials in echolocating bats. Journal of the Acoustical Society of America, 158(4): 3390-3399. https://doi.org/10.1121/10.0039659
Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality
Anya is LIVE right now
FREE
Free to watch • No registration required • HD streaming
Alright gang, I started organizing my To Read papers for the summer and uhhhhh they're like 98% auditory neuroscience. Because that's my whole deal. BUT I figured that would be boring, so here's a poll to determine what other topics I'll try to cover at least here and there:
What other topics would you like to see me cover at least once?
The second annual Reading-Neuro Summertime Extravaganza is nearly upon us!
For the uninitiated, last summer I tried my best to read a scientific article ever work day and upload a summary here. With a smattering of failures, as is to be expected with any new projects. And here we are, on the cusp of another summer! However, a caveat: Comprehensive Exams.
I am in the portion of my PhD training where I basically have to prove to a panel of faculty that I am a good enough scientist to continue on in this doctoral program. Naturally, this is a bit intimidating. It also means that I have to draw a sharper focus on my research to make it the best it can be, so I will have to spend more time on my lab work and less time reading articles about How Crocodiles Perceive Human Babies Crying.
My current plan is to attempt to post three article summaries per week this summer. Expect a smattering (or more) of failures. These papers will also probably be much more auditory/bat focused, since my current To Read list is overrun with my favorite rabid little echolocators.
That being said, if you have any papers or topics you want me to check out, feel free to let me know!! I currently have 25 articles picked out, which may or may not be enough for the summer.
Let's do this!! (<- excitement, because the other option is Fear, which I don't find nearly as fun)
"i look forward to hearing back" implies a beautiful world that runs on sense-direction combinations. i smell sideways to tasting up. i palpate inwards to listening diagonal, so that i can hunger clockwise
yeah it's been a minute, as always, so here's a paper I read for fun recently:
Neural Response Reliability as a Marker of the Transition of Neural Codes along Auditory Pathways by Buck et al. (2025).
What a fancy-shmancy paper!!
A quick briefing on the auditory system: sound starts in your ear, gets turned into electrical signals in your cochlea, goes to your brain through the auditory nerve, gets bounced around a bunch in your brainstem, then through the inferior colliculus (midbrain structure), then through the thalamus (SUPER important sensory gate-- all kinds of stuff goes through here), then finally (ish) to your auditory cortex. Then to secondary auditory structures BUT THAT'S NOT RELEVANT RIGHT NOW. Point is: complicated pathway, and each step does something slightly different with the information.
This paper basically took recordings from a couple auditory areas and checked how well they could decode the neural signals to understand what the original sound was. To be clear: this is crazy. We are getting better and better at starting with brain signals and working backwards, because we're getting better and better at understanding how those signals are made in the first place.
They specifically found that the general mechanism for encoding sound information changes as you go along the pathway. In the beginning steps, the electrical signals use temporal coding, meaning the specific pattern of neurons firing over time contains information about the sound. However, in later stages, brain regions start to use more rate coding, meaning the information isn't stored in a pattern of firing but just in how fast the neurons are firing. It's the difference between Morse code (temporal coding-- the pattern holds the information) and a movie's framerate (rate coding-- the speed is what matters most).
Super cool!
Love ya <3
Buck, A., Dupont, T., Andrews Cavanagh, R., Postal, O., Bourien, J., Puel, J.L., Michalski, N., & Gourévitch, B. (2025). Neural Response Reliability as a Marker of the Transition of Neural Codes along Auditory Pathways. Adv Sci (Weinh). 2025 Dec;12(46):e08777. doi: 10.1002/advs.202508777.
this is so interesting, because I have OCD and an autoimmune disease that causes inflammation, and those things are 100% related. when my immunosuppressant meds are working, my OCD stops. and when I’m unmedicated it’s like my brain is on fire. the one time I actually saw a doctor for my OCD and went through multiple sessions for it, it was allllll about breaking habits and learning self control, and would you believe it, it didn’t help one bit.
Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality
Anya is LIVE right now
FREE
Free to watch • No registration required • HD streaming
blakekathryn
janainaart
origamiaround
stormborns
turnnoffyourmind
titsay
izzy806
andulkaphoto
tangentwithtanner
