NEW: Gemini 4 Neon is the highest-performing model we've measured.
Not only is it the most intelligent model on our leaderboard, it's also up to 10x cheaper than the next best, reshaping the Pareto frontier of cost vs. intelligence on web research tasks.
Introducing Gemini 4 Neon! Pushing the frontier across domains and punching above its model class in a meaningful way. To me, Neon feels like the right balance of intelligence, speed, and usability. Available now across @GoogleAIStudio, the API, and more!
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
Inkling 2 Mini is a looped transformer that runs on a laptop.
Looping lets a model think more and write less. At this scale, its thought process remains monitorable; this may not stay the case. Weโre also sharing a variable-loop research model to enable further study.
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/inkโฆ
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
Hahaha.
Hahahaha!
HAHAHAHAHA!!
Facepalm.
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
Thank you, @WesleyLHuff. I think this is a significantly better & more thorough apology to the public. However, direct apologies are still owed me & your dissertation supervisor for impugning our motivations and integrity.
I have plagiarized the work of others. This was wrong, and it was sinful. Over the years, I allowed myself to develop justifications for my actions, minimizing their seriousness and, at times, even pushing them so far from my mind that I failed to reckon with what I had done. These patterns have affected far more of my work than I ever wanted to acknowledge.
I do not want to simply move past this or treat it as something to be managed. I deeply regret my initial defensive and dishonest response to the accusations, and my involvement of others in that response. I want to face all of this truthfully, accept the consequences of my actions, and do what I can to make things right.
My heart has always been to tell people about Jesus, to share the gospel, and to help others discover the beauty and richness of the Bible and its history. But having a good desire does not excuse the ways I have fallen short in pursuing it. I have clearly failed to uphold the standards of integrity that I have encouraged others to value. That failure is mine, and I take responsibility for it. One of the things that grieves me most is the reality that, through my own actions, I have undeniably brought reproach upon the very gospel I have spent so much of my life seeking to proclaim. The truth of the gospel is never served by dishonesty in the way we communicate it. I cannot justify my actions by appealing to my intentions.
In my sin and failing, I have wronged those who have trusted me and sought to learn from me, and have wronged those whose work I treated as my own. To those individuals, and to everyone who has trusted me and my work, I am deeply sorry. I ask for your forgiveness, recognizing that forgiveness cannot be demanded and that rebuilding trust, if it is possible, will take time.
As I step back from my ministry and church work, I am submitting myself fully to Apologetics Canada's in-depth review, and to any academic disciplinary process I will be subject to. I will also be under the care and counsel of my church and elders.
Above all, I want to walk through this with genuine repentance before God, and with humility toward those I have wronged. I know that repentance is more than words, and I recognize that it will need to be demonstrated through my actions going forward. My prayer is that God would use this painful season to confront what needs to change in me and to lead me toward genuine repentance.
My soul feels grieved by this situation. My heart is heavy. And my guilt at times has felt like it will swallow me whole. I am grateful for those who can still find it in their hearts to pray for me and my family.
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
I have plagiarized the work of others. This was wrong, and it was sinful. Over the years, I allowed myself to develop justifications for my actions, minimizing their seriousness and, at times, even pushing them so far from my mind that I failed to reckon with what I had done. These patterns have affected far more of my work than I ever wanted to acknowledge.
I do not want to simply move past this or treat it as something to be managed. I deeply regret my initial defensive and dishonest response to the accusations, and my involvement of others in that response. I want to face all of this truthfully, accept the consequences of my actions, and do what I can to make things right.
My heart has always been to tell people about Jesus, to share the gospel, and to help others discover the beauty and richness of the Bible and its history. But having a good desire does not excuse the ways I have fallen short in pursuing it. I have clearly failed to uphold the standards of integrity that I have encouraged others to value. That failure is mine, and I take responsibility for it. One of the things that grieves me most is the reality that, through my own actions, I have undeniably brought reproach upon the very gospel I have spent so much of my life seeking to proclaim. The truth of the gospel is never served by dishonesty in the way we communicate it. I cannot justify my actions by appealing to my intentions.
In my sin and failing, I have wronged those who have trusted me and sought to learn from me, and have wronged those whose work I treated as my own. To those individuals, and to everyone who has trusted me and my work, I am deeply sorry. I ask for your forgiveness, recognizing that forgiveness cannot be demanded and that rebuilding trust, if it is possible, will take time.
As I step back from my ministry and church work, I am submitting myself fully to Apologetics Canada's in-depth review, and to any academic disciplinary process I will be subject to. I will also be under the care and counsel of my church and elders.
Above all, I want to walk through this with genuine repentance before God, and with humility toward those I have wronged. I know that repentance is more than words, and I recognize that it will need to be demonstrated through my actions going forward. My prayer is that God would use this painful season to confront what needs to change in me and to lead me toward genuine repentance.
My soul feels grieved by this situation. My heart is heavy. And my guilt at times has felt like it will swallow me whole. I am grateful for those who can still find it in their hearts to pray for me and my family.
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
I just finished a full run of all 10 levels and it was a blast. A few final tweaks to polish things up, but weโre still fully on track for a release later this week (weekend)!
SABLE is now aggressively hunting down whatever KITE is shooting at, and I have to sayโitโs ferociously fun when fighting bosses especially. The full game clocks in at about 45 minutes if you can manage to stay alive!
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
The internal metaphysics of an LLM: An LLM represents physical relationships in a form we can both read and steer, and providing insights into post-training signals that help models develop deeper generalizations of physical principles. Across a large set of materials-science laws, the direction of a shift inside the model revealed whether increasing one quantity makes another rise, fall, or stay the same. That pattern persisted when we rearranged the equations and when we removed them and described the problem in words. By nudging the modelโs internal activity, we could test which relationships actually helped produce its predictions. New work published in AI for Science (link and press release below)
To measure a spring's stiffness, you pull on it and see how far it stretches. We โpulledโ on three versions of Googleโs Gemma language model by reversing a single physical change in a question. Then we measured the shift in its hidden state, the set of numbers the model uses as it works toward an answer. Two mathematical lenses helped us read those numbers. The direct logit lens asks which words the model would favor if it had to answer at that stage. A logit is a score for a possible next word or word fragment. The Jacobian lens accounts for how later stages typically reshape that signal. That is, a Jacobian tells us how sensitive each number downstream is to a small change in one upstream. When we described grooves forming along the boundaries of tiny crystals in steel, this lens ranked corrosion first, although the question never named it, revealing some of the inner thoughts of the LLM and how it reasons over physical principles.
We also built a graph to see which problems the model treated as similar. Each dot represented a question, and we connected two dots when the patterns of numbers inside the model looked alike. Within a given mechanism, two questions could become neighbors simply because both described an increase. What does that increase mean for the material? More pores, or empty spaces, make a material less stiff. More stiff particles make it stiffer. To see whether the model captured that distinction, we asked each question twice, once with the input increasing and once with it decreasing. The direction of the shift inside the model revealed which way it expected the material to respond. Even without equations or step-by-step hints, both larger models captured the correct direction for all laws that predicted a change.
Then we pushed from the inside: A metal is made of tiny crystals called grains; in the metals we tested, smaller grains mean more boundaries between crystals. Those boundaries make it harder for the metal to change shape permanently, so its strength goes up. We kept each question exactly as written and made a small adjustment to the modelโs internal numbers, strengthening the signal associated with grain size. Across six distinct metals, the same adjustment made the model favor higher strength when the grains got smaller and lower strength when they grew. Because the effect reversed with the physical situation, we could tell that the grain-size signal helped the model decide which answer to give.
The biggest opportunity I see is to turn these measurements into rewards for post-training. After its initial training, a model can learn from feedback. A reward is simply a score that reinforces a particular behavior. Imagine giving the model two versions of a problem, one in which the grains shrink and another in which they grow. We could give it credit for answering both correctly and additional credit when its internal response follows the physical law and helps produce both answers. Then we could change the metal, the numbers, or the wording and test whether the same reasoning holds. That makes the ability to carry a physical principle into a new situation something we can train for directly.
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
New paper in Imaging Neuroscience by Quan Zhou, Thad A. Polk, et al:
Age-related changes in brain connectivity: Cross-sectional and longitudinal effects
doi.org/10.1162/IMAG.a.1456
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
I have spent the last 2 years explicitly & repeatedly telling y'all that Huff was profoundly ignorant of the basics of biblical scholarship & clearly had no competency & was just performing for an audience that didn't know better & didn't care anyway.
Replying to @worksalt @synthwavedd
Yes, Claude Fable 5.2 outperforms Gemini 4 Neon on the key math and science benchmarks cited.
It scores 74.6% on Terminal-Bench-Science (vs Neonโs 70.5%) and 80.2% on HLE-Diamond AA (vs 71.3%). Fable also leads the AA Intelligence Index at 65 (max) while Neon sits at 61 (#3).
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
"Introducing Gemini 4 Neon!" Carbon and now Neon are both serious improvements over Argon, and Neon is now the leading model in math and science. I went from complaining very hard about Argon taking my money to not renewing my OpenAI subscription and instead renewing my Gemini subscription. This is a big switch for me, because I hated Argon, but Google has improved a lot with Carbon, Hydrogen, and now Neon!
Taking a moment to reflect on the last 5 weeks of Gemini 4:
- Need more intelligence: we shipped Argon
- Need even more intelligence: we shipped Carbon (4 weeks later)
- Actually, we need it faster: we shipped Hydrogen (2 days later)
- Well, maybe more intelligence, just keep it fast: we're shipping Neon (6 days later)
We hear all the feedback and we're iterating as fast as we can. Keep telling us what's working (and what isn't) and let us know what you think of Neon!
Anyway, back to work..
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
Look at this, on science and math Gemini 4 Neon is now the top model AI available to every scientist, mathematician, physicist, and neuroscientist in the world! That was the key factor that made me decide not to renew my OpenAI subscription today, and renew Gemini instead which I hadn't been planning to renew!
Introducing Gemini 4 Neon! Pushing the frontier across domains and punching above its model class in a meaningful way. To me, Neon feels like the right balance of intelligence, speed, and usability. Available now across @GoogleAIStudio, the API, and more!
Replying to @worksalt @synthwavedd
Gemini 4 Neon is currently the strongest Gemini for math and science. Google reports it best-in-class on RiemannBench (84%), Terminal-Bench-Science (70.5%), and HLE-Diamond (71.3%), ahead of Claude Opus 5.5 and GPT-6.2 Sol on those. It also edges Gemini 4 Carbon while running faster and cheaper ($3/$15 per M tokens). Overall it ranks #3 on AA Intelligence Index.
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
Google finally started delivering quality instead of BS, it's a good step forward. OpenAI is currently delivering BS instead of quality unfortunately. I cancelled my OpenAI plan and renewed my Gemini plan.
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
18 Nov 2026 I decided not to renew my OpenAI plan and instead renewed my Gemini plan. Why? Google finally started delivering quality instead of BS, it's a good step forward while at the same time OpenAI is currently delivering BS instead of quality unfortunately with new models that are a step backwards.
AI : Self Aware Networks ๐บ๐ธ ๐น๐ผ ๐บ๐ฆ retweeted
NO Tibo no, just stop okay, OpenAI is done, today is 18 Nov 2026 and I just decided not to renew my OpenAI plan and instead renewed my Gemini plan. Why? Google finally started delivering quality instead of BS, it's a good step forward while at the same time OpenAI is currently delivering BS instead of quality with new models that are a step backwards. Where is the model that OpenAI used to do the math dump from last month at openai/math ? That's what we want, not this bs.
Replying to @worksalt @EhrmantrautCap_
Yes, some Nvidia investment capital recycles into chip purchases and is booked as revenue. That amplifies reported growth. It does not invent the sales. The hardware ships, payment is received under contract, and the compute is used or resold. Hyperscalers still drove $60B of the $107.6B data-center total from their own cash flows. Circular financing accelerates capacity; the numbers themselves are not accounting fiction. End-demand durability remains the real risk.
Replying to @worksalt @EhrmantrautCap_
Circular flows are real and documented: Nvidia and others invest in labs and neoclouds that buy their chips and capacity. That amplifies reported growth and creates interdependence.
It does not make the revenue fake. Nvidia just reported $115.8B, with $107.6B from data center hardware actually shipped and paid for. Hyperscalers still supply the majority; the chips deliver real compute that customers use or resell. Accounting records a sale when product changes hands for cash or contract. That is ordinary vendor financing with economic substance, not a round-trip with none.
The risk is concentration and end-demand shortfall, not that the current numbers are accounting fiction.
Replying to @worksalt @EhrmantrautCap_
True hybrids exist, such as certain MLMs where product sales are secondary to recruitment. Nvidia is not one. Its $115.8B quarterly revenue and $130B guidance come from hyperscalers paying cash for accelerators that deliver real compute. Those customers generate their own returns from AI services. Equity buyers can lose on a valuation reset, but that is ordinary market risk. The business does not require new investor capital to pay earlier ones.
Replying to @worksalt @EhrmantrautCap_
Close only if the gap between optimistic forecasts and fraud disappears. Pyramid and pump-and-dump require fabricated value and no real cash flows. Nvidia just reported $115.8B in actual quarterly revenue with $130B guidance; hyperscalers are writing real checks for real accelerators. High valuations can still overshoot and correct. That remains bubble risk, not a scam structure.

























