The Pawnisher retweeted
Hiroo Isono (1945–2013) was a Japanese illustrator known for his dreamlike landscapes and lush scenes of fantastical nature. His work blends elements of fantasy, science fiction, and ecology, portraying worlds where vegetation and wildlife dominate over human presence.
His highly detailed and vibrant style also left a mark on the video game world: he created promotional illustrations for Secret of Mana, the classic RPG by Square released in 1993. His images helped define the game’s magical and nature-driven atmosphere, becoming an iconic part of its visual identity. 🌿🎨
The Pawnisher retweeted
Un recuento de las intervenciones de Estados Unidos en América Latina que llenaron el continente de sangre.
The Pawnisher retweeted
El robot celeste farmeando aura y El rojo pasando la verguenza de su vida
A ese lo van a reciclar
The Pawnisher retweeted
"Probability is the logic of science."
There is a deep truth behind this conventional wisdom: probability is the mathematical extension of logic, augmenting our reasoning toolkit with the concept of uncertainty.
In-depth exploration of probabilistic thinking incoming.
The Pawnisher retweeted
This man solved one of the hardest puzzles of the century.
Rejected his $1 million award and vanished into Russia.
Everyone called him crazy.
Here's the forgotten story of Grigori Perelman—the craziest genius mathematician everyone should know: 🧵
The Pawnisher retweeted
❤️ Lo más bello (y curioso) que vas a ver hoy:
Un chip de Apple (5nm) al microscopio electrónico.
Miles de millones de transistores ordenados por bloques.
The Pawnisher retweeted
Muchos aspectos de la física pueden ser pensados en términos de formas diferenciales, no obstante a menudo no tenemos mucha idea del trasfondo geométrico de estas.
Comienzo así una serie de hilos donde trataremos de ver la física mediante formas 😳😳 (finalmente).
The Pawnisher retweeted
Me trying to hide from my problems. teapawt.net/majeliskucing/status/1…
The Pawnisher retweeted
“As Leibniz suggested, we appear to live in the best of all possible worlds, where the computable functions make life predictable enough to be survivable, while the uncomputable functions make life unpredictable enough to remain interesting"
ift.tt/120jfCP
The Pawnisher retweeted
Do the RAG GenAI systems of the big law tech companies give the “hallucination-free” drafting assistance they claim?
@chrmanning & Mirac Suzgun aided a @Stanford RegLab study showing they do much better than GPT-4 but still hallucinate about 1 time in 6.
hai.stanford.edu/news/ai-tri…
The Pawnisher retweeted
Caspar David Friedrich, el pintor romántico más conocido. De él nos vienen a la mente "El caminante sobre el mar de nubes" y "La abadía del robledal". Pero el artista tiene otras muchas obras menos conocidas pero sobrecogedoras.
🧵En este hilo tenéis algunas de mis favoritas.
The Pawnisher retweeted
Retrieval-augmented generation (RAG) is the best way to specialize an LLM over your own data. Researchers have recently discovered a finetuning approach that makes LLMs much better at RAG...
RAFT and specializing LLMs. Most use cases with LLMs require specializing the model to understand new information that was not present during pretraining or finetuning. For example, we might have a set of company documents about which we want the model to answer questions. What is the best way to bake this knowledge into our LLM? This question has been studied a lot recently, spurring the creation of a research topic referred to as knowledge injection. Within this line of research, we have seen that retrieval augmented generation (RAG) is an effective approach [2], but authors in [1] extend this technique by proposing retrieval augmented finetuning (RAFT)—a finetuning recipe for making LLMs better at RAG.
Different QA setups. When performing RAG, we assume access to information that can be used by the LLM to better answer a question. However, there are a few different problem setups that are commonly considered for performing question answering with an LLM:
- Closed book: use the LLM as a chatbot and rely upon its internal (parametric) knowledge to answer the question properly.
- Open book: allow the LLM to refer to external sources of information1 when answering the question.
In [1], we study a modification of the open book setup called domain-specific open book question answering. This setup assumes that we have both i) an open book setup and ii) prior knowledge of the domain in which the LLM will be tested.
Benefit of domain-sepcific QA. By having access to relevant, domain-specific information ahead of time, we can finetune the model on this data to aid the knowledge injection process. For example, if we want to build an LLM to answer questions based on a fixed set of internal company documents, we can finetune our model on these documents prior to performing any question answering. As we will see, RAFT is just a particular finetuning strategy that is proposed for domain-specific open book question answering tasks. This finetuning strategy focuses upon improving the LLM’s ability to perform RAG, thus boosting the quality of knowledge injection.
What is RAFT? The first step in performing RAFT is to construct a training dataset. Each data example in this training dataset has three components:
1. A question that the LLM is expected to answer.
2. A set of documents that the LLM can use to help answer the question.
3. A chain-of-thought (CoT)-style answer generated from one of the documents.
Using these components, we can use a prompt template to construct training examples with the structure shown in the image below. Here, we should notice that portions of the response that are directly sampled from a document are surrounded by the special ##begin_quote## and ##end_quote## tokens, which allow the model to learn to rely upon document context when producing an answer.
We can then finetune the LLM—using supervised finetuning (SFT)—on such data to produce answers from the provided documents. When creating training examples for RAFT, we use CoT-style responses (i.e., these are usually generated with the help of GPT-4) in particular to hone the model’s reasoning capabilities. The model learns to output CoT-style answers based upon the question and provided context. This finetuning process is the core of the RAFT approach.
Document types. Although each training example in RAFT contains a set of documents, we can further classify these documents as one of two types:
- Oracle documents: the answer to the question can be reasonably deduced from these documents.
- Distractor documents: these documents do not contain useful (answer-related) information.
During training with RAFT, we set a hyperparameter P, such that P% of questions in the dataset have both oracle and discriminator documents included within the training example. For 1-P% of the documents, we only use distractor documents and remove the oracle document. By doing this, RAFT trains the model to disregard documents that have no useful information for crafting a response.
“By removing the oracle documents in some instances of the training data, we are compelling the model to memorize domain-knowledge.” - from [1]
Additionally, excluding relevant documents also forces the LLM to learn new, domain-specific information during finetuning to a a certain extent. See below for a depiction of the RAFT pipeline and its comparison to standard RAG.
Comparison to prior work. The RAFT approach is very similar to finetuning strategies proposed for teaching LLMs to make valid API calls; e.g., Gorilla or RAT. However, RAFT generalizes these approaches beyond just API calls and focuses on making the model generally better at using any form of context for RAG applications.
Empirical evaluation. In [1], authors finetune a LLaMA-2-7B model using the RAFT approach. When evaluating this model, we see that it becomes much better at performing RAG on the documents upon which it is trained; see the image below. Although these results are exactly what we would expect, they truly demonstrate the benefit of RAFT for domain-specific open book question answering. If we have access to the documents about which we want an LLM to answer questions ahead of time, then we can finetune the LLM over these documents (using RAFT) to significantly improve the quality of RAG!
-----
Bibliography
[1] Zhang, Tianjun, et al. "RAFT: Adapting Language Model to Domain Specific RAG.” arXiv preprint arXiv:2403.10131 (2024).
[2] Ovadia, Oded, et al. "Fine-tuning or retrieval? comparing knowledge injection in llms." arXiv preprint arXiv:2312.05934 (2023).
The Pawnisher retweeted
Hablemos de Hitoshi Ashinano 🧵
Un mangaka particular que hace obras Slice of Life, para ser más exacto tirando a lo Iyashikei: obras que se centran en proporcionar una sensación de calma y tranquilidad a sus consumidores.
Él exhibe ciertos rasgos recurrentes en sus obras.
The Pawnisher retweeted
Solving a maze using front propagation (Fast-Marching method).
en.wikipedia.org/wiki/Fast_m…
The Pawnisher retweeted
¿Qué es la eternidad? La eternidad no es la suma de todos nuestros ayeres. La eternidad es todos nuestros ayeres, todos los ayeres de todos los seres conscientes. Todo el pasado, ese pasado que no se sabe cuándo empezó. Y luego, todo el presente. Este momento presente que abarca todas las ciudades, todos los mundos, el espacio entre los planetas. Y luego, el porvenir. El porvenir, que no ha sido creado aún, pero que también existe. Los teólogos suponen que la eternidad viene a ser un instante en el cual se juntan milagrosamente esos diversos tiempos. Podemos usar las palabras de Plotino, que sintió profundamente el problema del tiempo. Plotino dice: hay tres tiempos, y los tres son el presente. Uno es el presente actual, el momento en que hablo. Es decir, el momento en que hablé, porque ya ese momento pertenece al pasado. Y luego tenemos el otro, que es el presente del pasado, que se llama memoria. Y el otro, el presente del porvenir, que viene a ser lo que imaginan nuestra esperanza o nuestro miedo.
Jorge Luis Borges.
"Borges oral" (cinco conferencias pronunciadas en la Universidad de Belgrano, recopiladas por Martín Müller). Emecé. Buenos Aires, 1979.

































