Jhonathan P. R. dos Santos (@jhowpd) header
Jhonathan P. R. dos Santos (@jhowpd) profile picture

Research Scientist at #Corteva (tweets are my own) | #Genetics, #PlantBreeding, #Analytics, #AI

Johnston, IA
Joined March 2018
Photo by Jhonathan P. R. dos Santos (@jhowpd)
Pro tip when playing with pybind11: if creating a custom optional type, make sure to use std::variant instead of std::optional, and create an empty struct to mimic the None from python. It will be seemless python, with has_value, value_or, set_value reset, and repr (basic) types.
82
why is so hard for openai create a context setup that once you asked only for giving code with no explanation, just remember this. How annoying is that.
1
1
119
Jhonathan P. R. dos Santos retweeted
.@ggreenwald highlights @elonmusk's opposition to Brazil's authoritarian censorship regime. The censorship framework in Brazil mirrors tactics employed by the US government to suppress dissenting voices under the guise of combating disinformation. "Any criticism of this judge that offends him will result in punishment, no trial, no due process, and that's what they've done now to Elon Musk." "This censorship regime has no basis in law. They actually tried to pass a law several months ago that would democratically enact this censorship regime to formally define the powers and the responsibilities of big tech. But they could not get the votes in Congress to pass it." Thus, Judge Alexandre de Moraes was given unilateral authority to dictate permissible online speech. "When they issue these orders to these platforms, they say, we want this member of Congress banned, we want this journalist banned, we want this activist banned. They provide no due process. They provide no explanation. And these orders are never sent to the individuals who are being censored." "There is no judicial process for contesting the order. If you do try to contest it, you go right to the judge who issued it, and obviously, you have no chance of being heard." "It's not just censorship. It's censorship with no due process in the most Kafkaesque way; there's not even a legal framework for it." "They're not even allowed to inform their users that they received a judicial order. Let alone disclose to the public or their users what the content of the order is. They just have to obey and ban the individuals named in the order, and if they don't do so within two hours of receipt, they will face daily fines of $20,000 for every account they fail to ban." In January 2023, Greenwald obtained one of these judicial orders that revealed arbitrary bans on democratically elected congress members of the House and Senate, including 25-year-old politician Nikolas Ferreira, who garnered record-breaking support. Ferreira not only secured a historic victory by amassing the largest vote total of any congressional candidate in Brazil's history but also garnered over a million votes in his state, surpassing all other candidates in the 2022 election. However, despite his widespread support, the judge overseeing the censorship regime arbitrarily banned Ferreira from all online platforms, silenced him, and deprived him of his online presence without any explanation or opportunity for due process. Senators, journalists, and activists who express dissenting opinions or criticize this censorship are similarly targeted for censorship by the same judge, highlighting the authoritarian nature of these actions. "Needless to say, Google, YouTube, Facebook, Instagram, and WhatsApp, which is by far the most popular chat platform in Brazil, have always just meekly complied with every censorship order they got from Brazil, but Elon Musk decided he had enough."
1,225
6,645
539
26,203
13,405,582
Hack of the day: pass a list[int] from python as a binding with pybind11 and make it work in C++ without memory duplication and with int16 type, with seamless interaction from python. Lot of pain, but worked ;)
63
kind of funny how GPT-4 holds back explanations about CUDA Graphs.
69
Jhonathan P. R. dos Santos retweeted
Let me clear a *huge* misunderstanding here. The generation of mostly realistic-looking videos from prompts *does not* indicate that a system understands the physical world. Generation is very different from causal prediction from a world model. The space of plausible videos is very large, and a video generation system merely needs to produce *one* sample to succeed. The space of plausible continuations of a real video is *much* smaller, and generating a representative chunk of those is a much harder task, particularly when conditioned on an action. Furthermore, generating those continuations would be not only expensive but totally pointless. It's much more desirable to generate *abstract representations* of those continuations that eliminate details in the scene that are irrelevant to any action we might want to take. That is the whole point behind the JEPA (Joint Embedding Predictive Architecture), which is *not generative* and makes predictions in representation space. Our work on VICReg, I-JEPA, V-JEPA, and the works of others show that Joint Embedding architectures produce much better representations of visual inputs than generative architectures that reconstruct pixels (such as Variational AE, Masked AE, Denoising AE, etc). When using the learned representations as inputs to a supervised head trained on downstream tasks (without fine tuning the backbone), Joint Embedding beats generative. See the results table from the V-JEPA blog post or paper: ai.meta.com/blog/v-jepa-yann…
This Post is from an account that no longer exists. Learn more
182
715
154
4,716
2,009,170
Programming is akin to cleaning: Start somewhere, address the nearest issue, and progress to the next, aiming for utmost cleanliness. In coding, substitute 'dirt' with 'errors', you learned how to code.
2
81
Jhonathan P. R. dos Santos retweeted
It would be a useful exercise to imagine where the AI industry would be today if AI research labs in industry had remained closed, had not open sourced their code, had patented everything and enforced their patents. Imagine a world without PyTorch, and with enforced patents on Transformers, ResNet, Mask-RCNN, FPN, SAM, DINO, seq2seq, wav2vec, memory-augmented networks, Tansformers, BatchNorm, LayerNorm, Adam, denoising auto-encoder, joint-embedding architectures, and tons of SSL methods....
AI is nothing without open source, #keepaiopen 🤗
106
412
58
2,703
743,476
Jhonathan P. R. dos Santos retweeted
This is a great little book to take you from “vaguely understand neural networks” to the modern broad state of practice. I saw very little to quibble with. fleuret.org/francois/lbdl.ht…
63
335
24
2,746
443,717
When I was in my second year of my Ph.D. one of the best decisions I ever made was migrating from R to Python. Today was a nice reminder of why I migrated out of R, 10 minutes to install the R base + one package to fit linear mixed models (lme4).
1
1
186
some advice if you are playing with bazel + rust + cargo raze + protobufs, make things flat, then will work.
104
wanna know true pain? try to setup bazel + rust + cargo raze + protobufs
131
Jhonathan P. R. dos Santos retweeted
Job alert! We are seeking a driven postdoc to explore phenomic selection as part of the #SweetCAP project. Apply now at apply.interfolio.com/121117 to join our team and contribute to the future of sweet corn! Please RT 🧬🌽🧬 #postdoc #postdocjobs #plantbreedingjobs
1
25
2
47
12,178
Jhonathan P. R. dos Santos retweeted
Stanford has released Prof. Manning's NLP course for free (!!!) on Youtube if you are interested in diving deeper into NLP. It starts with the very basic word2vec and expends to: -> Domain adaptation for supervised sentiment -> Retrieval augmented in-context learning -> Advanced behavioral evaluation -> Analysis methods -> More NLP methods Happy A.B.L <Always Be Learning> y'all! youtube.com/watch?v=PSGIodTN…
9
333
1
1,645
271,658
Interesting use of self-attention, modeling straight into long byte sequence: arxiv.org/pdf/2305.07185.pdf
56
please @excalidraw give us vim key bindings, and make it even more lovely 🩷
1
109
Jhonathan P. R. dos Santos retweeted
New paper on @TheorApplGenet "Leveraging probability concepts for cultivar recommendation in multi‑environment trials" @KaioOlimpio @matdkrause @jhowpd rdcu.be/cHy0F
1
6
28
Jhonathan P. R. dos Santos retweeted
Excited to share our latest paper published by @PNASNews. By combining metabolomics and machine learning methods we predicted consumer acceptability and provided new tools for flavor-assisted selection in plant breeding. Link: pnas.org/content/119/7/e2115…
2
31
2
131
Jhonathan P. R. dos Santos retweeted
RAIL will be presenting a number of exciting late breaking poster results at the RL4RealLife WS #ICML2021 (8 pm PT today!): sites.google.com/view/RL4Rea… Algorithms for real-world RL w/ mobile manipulators, lifelong meta-learning methods, principled multi-task data sharing. A thread:
2
10
54
Recently I have been going a lot through papers, talks, online classes from @svlevine, outstanding material and good science, I definitely recommend for anyone interested about machine learning and AI in general.
3