REVIEW 4 major objections 6 minor 2 cited by
On Path to Multimodal Generalist: General-Level and General-Bench
T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read A five-level synergy test finds most AI generalists fall short
desk verdict A large, genuinely useful multimodal benchmark whose 'synergy' leaderboard rests on an admitted and unvalidated equivalence; worth refereeing, but not as-is. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the General-Level scoring ladder. Level-2 averages normalized scores over all supported comprehension and generation tasks; Level-3 re-scores each task as zero unless the generalist beats the task's SoTA specialist, so the score counts only "winning" tasks; Level-4 takes the harmonic mean of comprehension and generation scores, rewarding balance; Level-5 multiplies the Level-4 score by a normalized weight equal to the model's rate of beating NLP SoTA specialists. Because higher levels are built from masked or combined lower-level scores, the framework mathematically guarantees scores decrease monotonically as levels rise. The synergy concept is what carries the argument: the paper treats beating a specialist as observable evidence of transfer, and General-Bench supplies the task surface—702 tasks in native formats, grouped by modality and by comprehension and generation—on which that evidence is collected.
What would settle it
Train two copies of a top Level-3 model—one jointly on a set of tasks, one on each task independently—and compare their performance against the same specialists; if the separately trained copy matches or beats the joint model, then the measured "synergy" is not transfer and the Level-3 and higher hierarchy loses its evidential basis.
Extended reading notes
Core claim
On the paper's own terms, the central discovery is that "synergy"—defined as a generalist outperforming the state-of-the-art (SoTA) specialist on a task, taken as evidence of transfer from other tasks or modalities—is rare and shallow in current MLLMs. At Level-2 (basic unified comprehension and generation), models like Unified-io-2 and AnyGPT outrank GPT-4V and GPT-4o because breadth of task and modality support outweighs single-task strength. At Level-3 (cross-task synergy), top ranks go to Sa2VA-26B, LLaVA-One-Vision-72B, and Qwen2-VL-72B, while GPT-4V and GPT-4o place lower. Only Mini-Gemini, Vitron-V1, and Emu2-37B reach Level-4, meaning synergy across comprehension and generation. No model earns a non-zero Level-5 score: no tested system outperforms NLP SoTA specialists on language tasks, so there is no evidence that non-language modalities enhance language intelligence.
Load-bearing premise
The whole measurement of synergy assumes that a generalist beating a chosen specialist on a task proves it transferred knowledge from other tasks or modalities, rather than simply being larger, better trained, or matched against weaker baselines.
Editorial extensions
If this is right
- Rankings of MLLMs change once synergy is the criterion: models that support many modalities and beat specialists on many tasks rise, while high-scoring but narrow models such as GPT-4V fall.
- Future training of multimodal generalists should explicitly target cross-task and cross-modal transfer, because Level-3 and Level-4 cannot be reached by adding parameters or data within a single task.
- The absence of any Level-5 model implies that current language-centric MLLM architectures are not yet producing bidirectional modality-to-language transfer, so improving that direction becomes a named research goal.
- Benchmarks should preserve native task formats rather than coercing everything into multiple choice, since forced QA hides generation and fine-grained-output failures.
- Because specialist baselines update over time, a generalist's level is not permanent: as SoTA specialists improve, models must keep improving to hold their Level-3 and higher status.
Reading between the lines
- Because "synergy" is measured against a chosen pool of open-source SoTA specialists, the same model could land at different levels if that pool is swapped; a natural test is to re-run the Level-3 and higher scoring against a stronger or weaker specialist pool and watch the rankings move.
- The framework's own monotonicity proofs imply that higher levels are definitionally harder to reach, meaning a model's level is only partly an empirical fact about the model and partly a design choice about task coverage and specialist baselines.
- The paper's observed image-video synergy clustering suggests visual modalities share transferable features; a testable implication is that joint image-video training should produce the largest specialist-beating gains, while audio-language joint training should show the smallest, guiding where to invest in architecture.
- The empty Level-5 predicts that simply adding more multimodal pretraining to an LLM will not improve its core NLP performance; this could be tested by measuring an MLLM's NLP scores before and after multimodal training under controlled data budgets.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes General-Level, a five-level taxonomy for ranking multimodal large language models (MLLMs) by a construct called 'synergy,' and General-Bench, a benchmark of 702 tasks and 325,876 instances spanning image, video, audio, 3D, and language modalities in native task formats. At Level-2 and above, scores are computed from task performance relative to selected SoTA specialists: a model is credited with synergy on a task when its zero-shot score reaches or exceeds the specialist's score. The authors evaluate over 100 LLM/MLLM systems, report leaderboards at Levels 2-4, and conclude that most MLLMs lack cross-task and cross-modal synergy, that even GPT-4V and GPT-4o do not rank at the top, and that no model has demonstrated enhancement of language intelligence through non-language modalities.
Significance. If the central measurement were valid, the paper would offer a substantively new way to evaluate generalists beyond raw accuracy, and its negative result about current MLLMs would be a notable challenge to prevailing benchmark narratives. The benchmark itself is a large and potentially useful resource: 702 tasks in original free-form formats, 172 specialist references, 102 evaluated generalists, and broad modality and domain coverage. The observation that closed models such as GPT-4V rank below several open models under a specialist-relative scoring rule is also interesting and falsifiable. However, the headline synergy claims rest entirely on an unvalidated equivalence between beating a selected specialist and exhibiting cross-task transfer, and the formal monotonicity scaffolding contains algebraic errors. The benchmark and leaderboards can survive a reformulation, but the current framing overstates what the data establish.
major comments (4)
- [§3.2.2, §5.4, §6] The central measurement assumption is unvalidated. The paper defines a task-level 'win' over a SoTA specialist as evidence of a synergy effect, and all Level-3, Level-4, and Level-5 scores, together with the headline conclusions ('most MLLMs lack synergy'; 'no model has demonstrated the ability to enhance language intelligence through non-language modalities'), are built on that equivalence. The comparison is asymmetric: generalists are evaluated zero-shot (§5.3), while specialists are fine-tuned and selected for top public performance (§5.1), and the specialist set excludes closed-source models. A zero-shot generalist failing to beat a fine-tuned specialist cannot distinguish 'no learned cross-task transfer' from 'a high specialist bar by construction,' and a win over a weak or contaminated selected baseline cannot by itself establish transfer. The authors acknowledge in §6 that the relaxation 'avoids a direct measurement of the synergy effect.' As written, the Level-3+ rankings and the 'no reverse synergy' result are not measurements of synergy as defined in §3.1.2. I ask for either a validation study (for example, controlled single-task versus multi-task training where ground-truth transfer is known, checking whether beating a specialist predicts transfer) or a re-framing of all Level-3+ claims as 'zero-shot specialist-relative performance' with the term synergy removed.
- [§3.2.3, Table 1] The monotonicity proofs contain algebraic errors and do not establish the claimed strict decline. In the proof of S4 ≤ S3, multiplying the displayed inequality by 4(SC + SG) yields (SC + SG)^3 ≥ 8 SC SG, not ≥ 8 SC SG (SC + SG); the subsequent factorization is not a consequence of the displayed inequality. The correct AM-HM argument does show S4 ≤ S3 for positive scores, but only as a non-strict inequality. Similarly, the proof of S3 ≤ S2 establishes ≤, and equality occurs whenever a model exceeds every specialist threshold. The paper's Property-2 claim that 'S_{k-1} > S_k' is therefore not proven. Additionally, Table 1 defines w_L = S_L / S_total without defining S_total, so the Level-5 score is not computable as specified. These issues do not necessarily invalidate the leaderboard numbers, but the formal support claimed in §3.2.3 should be corrected or downgraded to non-strict claims.
- [§4.1.2, §4.2, Table 2] The claimed minimum of 500 samples per task is inconsistent with the reported totals. Section 4.1.2 states 'We ensure that each task includes (at least) 500 data samples,' while Section 4.2 says 'For most of the tasks, we maintain around 500 testing instances.' Table 2 reports 702 tasks and 325,876 instances, an average of about 464 instances per task; the largest cell (271 image-comprehension tasks with 124,880 instances) averages about 461. These numbers are incompatible with a 500-sample minimum unless many tasks exceed 500 and others fall below. The paper should report the actual per-task distribution and reconcile the text. This matters because per-task specialist comparisons with small samples have high variance, which directly affects the stability of the Level-3 single-task 'win' decisions.
- [§5.1, §5.3, Tables 12-16] The leaderboard rankings are sensitive to the choice of the specialist baseline set, and no sensitivity analysis is reported. The paper excludes closed-source models from the specialist pool (§5.1) and selects specialists by public benchmark recognition, so the 'win-over-specialist' counts (Observation-2 in §5.3) and all Level-3+ scores are relative to this particular set. For example, GPT-4V and GPT-4o are evaluated as generalists but cannot serve as specialist references; had they been included for the tasks they support, several Level-3 wins could disappear. The paper should report how rankings change under alternative specialist choices (for example, best open-source model per task versus best available model, or including closed-source API baselines where feasible) before claiming that the rankings are stable characterizations of generalist capability.
minor comments (6)
- [§3.3] The section heading 'Receipt to Leveling Upper in General-Level' appears to be a typo; it should read 'Recipe for Leveling Up in General-Level.'
- [Table 4] The backbone name 'Qwev-7B' appears twice (models 10 and 22); this is presumably 'Qwen-7B' and should be corrected.
- [§5.3, Observation-2] Observation-2 says 'few models capable of surpassing the SoTA generalist,' but the comparison is against specialists; the terminology should be fixed to 'SoTA specialist.'
- [Tables 6-7 vs. Table 4] The model is named 'SEED-LLaMA-13B' in Table 4 but 'SEED-LLaMA-14B' in Tables 6 and 7; the naming should be consistent across the paper.
- [§4.3] The text says 'General-Bench comprises 130 multimodal skills, containing 702 tasks,' while Table 2 and Figure 7 report 145 skills; please clarify whether language skills are excluded from the 130 figure.
- [§3.2.3, Property-3] The 'More Task, The Better' argument is not generally true as stated: since S2 is an average over all benchmark tasks, adding a task where the model scores zero cannot increase its score, and a model supporting more tasks can still have a lower average if the added task scores are low. The claim should be reworded to describe an incentive under specific support/score conditions.
Circularity Check
Synergy is operationalized as beating a hand-picked SoTA specialist, so the paper's central 'no synergy' finding is a consequence of its own scoring definition.
-
self definitional
[Section 3.2.2 (Scoring Relaxation) and Section 5.3 Observation-2]
"Then, we can simplify the synergy measurement as: if a generalist outperforms a SoTA specialist in a specific task, we consider it as evidence of a synergy effect ... If an MLLM cannot outperform the SoTA specialist, it implies that the foundational conditions of cross-task/ability synergy for these MLLMs to become multimodal generalists are not met."
The paper defines evidence of synergy as exceeding a selected SoTA specialist's score (the mask=1 condition in S3's SC and SG), and then reports the absence of synergy whenever a model does not exceed that score. The central conclusion—'most MLLMs lack the cross-task or cross-modal synergy ability'—is therefore a restatement of the scoring rule, not an empirical discovery. The authors concede in §6 that this 'avoids a direct measurement of the synergy effect.' Because generalists are evaluated zero-shot (§5.3) while specialists are task-fine-tuned and chosen by the authors among open-source models (§5.1), the Level-3 and higher rankings measure beating a movable baseline, not learned cross-task transfer.
-
self definitional
[Section 3.2.1 (Scoring Specification, Level-5) and Section 5.6]
"Therefore, for level 5—measuring total synergy—we do not measure the generality across all modalities and tasks. Instead, we assess whether a model can improve NLP task performance to exceed that of NLP SoTA specialists."
The headline negative result, 'no model has yet demonstrated the ability to enhance language intelligence through non-language modalities,' is read directly from the Level-5 formula S5 = S4 × wL, where wL = SL/Stotal and SL is nonzero only when an NLP score σk ≥ σsota. Thus a model can only exhibit the claimed reverse modality-to-language synergy by exceeding fine-tuned NLP specialists; failing that threshold is, by construction, defined as lacking that synergy. The claim is therefore guaranteed by the scoring definition rather than measured as transfer. Additionally, Stotal is never defined, so the Level-5 score is formally underspecified even as a definition.
full rationale
The paper has a genuinely external component: General-Bench contributes 702 tasks and 325,800 instances with held-out evaluation, and the zero-shot performance numbers are not derived from the scoring definition. A model that actually exceeded many SoTA specialists on those tasks would receive a nonzero Level-3 or Level-4 score, so the benchmark is not vacuous as a capability report. However, the paper's load-bearing interpretive claim—that most MLLMs lack cross-task/cross-modal synergy, and that none shows non-language-to-language synergy—is made true by construction. Synergy is not measured; it is defined as surpassing an author-selected, open-source, task-fine-tuned SoTA specialist, while generalists are run zero-shot. The paper's own §6 limitation statement ('avoiding a direct measurement of the synergy effect') confirms this definitional gap. This is closer to a self-definitional reduction than to fitted-parameter circularity: the ranking tables and the 'no synergy' narrative are consequences of the threshold rule in §3.2.2, not independent findings about transfer. The self-citations to Vitron-V1 and other author-affiliated models are not load-bearing as a citation chain, and there is no imported uniqueness theorem; the circularity is concentrated in the operational definition of synergy itself. Separately, the mathematical scaffolding is unreliable (Stotal is undefined in S5, and the S4≤S3 proof contains an algebraic error), which weakens the formal framework but is not itself a circularity step. Overall, partial circularity at the core: score 6.
Assumptions & free parameters
free parameters (3)
- Metric normalization mapping functions =
Not specified in main text (Appendix A.1)
- Equal modality weighting 1/N =
1/4 for image, video, audio, 3D
- Stotal in Level-5 weight =
Undefined
assumptions (4)
- ad hoc to paper Outperforming a SoTA specialist on a task is treated as direct evidence of synergy
- domain assumption The selected open-source specialists represent the true SoTA ceilings for each task
- domain assumption Current MLLMs already transfer language intelligence to non-language modalities, so Level-5 only tests the reverse direction
- domain assumption Equal weighting of the four non-language modalities (1/N) produces an unbiased generality score
invented entities (2)
-
Synergy effect (as a measurable model property)
-
General-Level five-level taxonomy
Cite this review
Pith. "Pith review of On Path to Multimodal Generalist: General-Level and General-Bench." pith.science (2026). https://pith.science/paper/SYVEVWZM
@misc{pith2026250504620,
author = {Pith},
title = {Pith review of: On Path to Multimodal Generalist: General-Level and General-Bench},
year = {2026},
howpublished = {\url{https://pith.science/paper/SYVEVWZM}},
note = {Machine review of arXiv:2505.04620}
}
read the original abstract
The Multimodal Large Language Model (MLLM) is currently experiencing rapid growth, driven by the advanced capabilities of LLMs. Unlike earlier specialists, existing MLLMs are evolving towards a Multimodal Generalist paradigm. Initially limited to understanding multiple modalities, these models have advanced to not only comprehend but also generate across modalities. Their capabilities have expanded from coarse-grained to fine-grained multimodal understanding and from supporting limited modalities to arbitrary ones. While many benchmarks exist to assess MLLMs, a critical question arises: Can we simply assume that higher performance across tasks indicates a stronger MLLM capability, bringing us closer to human-level AI? We argue that the answer is not as straightforward as it seems. This project introduces General-Level, an evaluation framework that defines 5-scale levels of MLLM performance and generality, offering a methodology to compare MLLMs and gauge the progress of existing systems towards more robust multimodal generalists and, ultimately, towards AGI. At the core of the framework is the concept of Synergy, which measures whether models maintain consistent capabilities across comprehension and generation, and across multiple modalities. To support this evaluation, we present General-Bench, which encompasses a broader spectrum of skills, modalities, formats, and capabilities, including over 700 tasks and 325,800 instances. The evaluation results that involve over 100 existing state-of-the-art MLLMs uncover the capability rankings of generalists, highlighting the challenges in reaching genuine AI. We expect this project to pave the way for future research on next-generation multimodal foundation models, providing a robust infrastructure to accelerate the realization of AGI. Project page: https://generalist.top/
Figures
Figures from the paper (173 more)
Forward citations
Cited by 2 Pith papers
-
Dense360: Dense Understanding from Omnidirectional Panoramas
Introduces a 160K-panorama auto-annotated dataset, a dense captioning and grounding benchmark, and ERP-RoPE; fine-tuning Qwen2.5VL on the data lifts benchmark scores.
-
Mixed-R1: Unified Reward Perspective For Reasoning Capability in Multimodal Large Language Models
Mixed-R1 uses four reward types under GRPO, including a new bidirectional max-average token similarity (BMAS) reward, and lifts MLLM reasoning benchmarks by 2-5%.
Reference graph
Works this paper leans on
-
[2]
Worker for the audio group
-
[7]
多多参考本样例 Figure 73: Face Sketch Synthesis. 159 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: In-context Image Editing Paradigm: Generation Domain/Discipline: General-domain General/Universal Capability: Image Generation Input: Output: Task Name (Task Short Name) Image Editing with Text and Image Prompt (Img-Edit ...
-
[10]
多参考样例 Figure 107: Accent Classification. Skill (Meta-Task) Name: Speech Content Understanding Paradigm: Comprehension Domain/Discipline: General-domain General/Universal Capability: Reasoning Ability Task Name (Task Short Name) Intent Classification (Intent Recog) Requirement:
-
[13]
多参考样例 Input: Output: what is the intent in the audio? Action: decrease, Object: heat, Location: kitchen Figure 108: Intent Classification. 189 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Speech Emotion Understanding Paradigm: Comprehension Domain/Discipline: General-domain General/Universal Capability: Affecti...
-
[16]
多参考样例 Input: Output: what is the emotion in the audio? angry Figure 109: Speech Emotion Recognition. Skill (Meta-Task) Name: Music Understanding Paradigm: Comprehension Domain/Discipline: Art General/Universal Capability: Content Recognition，Commonsense Knowledge Task Name (Task Short Name) Music Instrument Classification(Instrument Recog) Requirement:
-
[19]
多参考样例 Input: Output: what is the instrument used in the music? brass Figure 110: Music Instrument Classification. 190 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Audio Technique Understanding Paradigm: Comprehension Domain/Discipline: Art General/Universal Capability: Reasoning Ability，Commonsense Knowledge Ta...
-
[22]
多参考样例 Input: Output: what vocal techniques are used in the given audio? Long tones, messa Figure 111: V ocal Technique Detection. Skill (Meta-Task) Name: Audio Content Understanding Paradigm: Comprehension Domain/Discipline: General-domain General/Universal Capability: Reasoning Ability，Content Recognition Task Name (Task Short Name) Long Audio Captioning...
-
[25]
"Air is moving through a large chamber and a loudspeaker is blaring close by
多参考样例 Input: Output: Listen to the audio and give a description. "Air is moving through a large chamber and a loudspeaker is blaring close by. Figure 112: Long Audio Captioning. 191 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: General Audio Question Answering Paradigm: Comprehension Domain/Discipline: General-d...
Show all 45 references
-
[28]
Figure 113: Open-ended Audio Question Answering
多参考样例 Input: Output: What can be inferred based on the combination of sound events in the clip? It is possible that the sound events are occurring at a social gathering or event where people are enjoying themselves and engaging in friendly conversation and laughter. Figure 113...
-
[31]
多参考样例 Input: Output: what is the animal sound in the audio? chicken Figure 114: Animal Sound Analysis. 192 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Environment Sound Understanding Paradigm: Comprehension Domain/Discipline: Gener...
-
[34]
193 On Path to Multimodal Generalist: General-Level and General-Bench Generation Tasks
多参考样例 Input: Output: what is the sound in the audio? drilling Figure 115: Sound Event Recognition. 193 On Path to Multimodal Generalist: General-Level and General-Bench Generation Tasks. Following we showcase 11 audio-oriented generative tasks, each representing a specific ski...
-
[37]
Figure 116: Audio Editing
多参考样例 edit the given audio file: This track composed of electronic instruments gives a sense of opening and clearness. Figure 116: Audio Editing. 194 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Dialogue Speech Generation Paradigm: ...
-
[40]
I'd like to try the chef's special
多参考样例 Dialog History: "I'd like to try the chef's special.", "Then what about you, Miss?", "I am on a diet. Do you have a vegetarian menu?", "The vegetarian dishes are at the last pages of the menu." " Oh, I see. I would like to have mashed potatoes and chocolate pudding. " Fi...
-
[43]
多参考样例 generate angry emotional speech from the given text, emotion and reference neural speech: 我刚从苏格兰回来 (I just came back from Scotland.)。 Figure 118: Emotional Speech Synthesis. 196 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Tex...
-
[46]
Figure 119: Multimodal Text-to-Speech (TTS)
多参考样例 generate a speech from the given reference image and text: Does society really exist as an entity over and above the agglomeration of men. Figure 119: Multimodal Text-to-Speech (TTS). 197 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task)...
-
[49]
多参考样例 generate audio from the given caption: A man speaking while a large crowd cheers in the background Figure 120: Single Captions To Audio Generation. 198 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Image-to-Audio Synthesis Para...
-
[52]
Figure 121: Image-to-Speech Synthesis
多参考样例 generate audio from the image: A blonde horse and a blonde girl in a black sweatshirt are staring at a fire in a barrel . Figure 121: Image-to-Speech Synthesis. 199 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Video-to-Audio S...
-
[55]
Figure 122: Video-to-Audio Synthesis
多参考样例 Generate an appropriate audio for the given video. Figure 122: Video-to-Audio Synthesis. 200 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Speech Style Transfer Paradigm: Generation Domain/Discipline: General-domain General/Uni...
-
[58]
Figure 123: V oice Conversion
多参考样例 generate male voice with Scottish accent to read the provided text: We would have concerns about suspension. Figure 123: V oice Conversion. Skill (Meta-Task) Name: Speech Translation Paradigm: Generation Domain/Discipline: Multi-domain General/Universal Capability: Conte...
-
[61]
有些地区也会有其他的人提供初级医疗，例如、药剂师。 Figure 124: Chinese-to-English Speech Translation
多参考样例 translate the given english sentence in to chinese speech: There are people providing primary care in some area, such as pharmacists. 有些地区也会有其他的人提供初级医疗，例如、药剂师。 Figure 124: Chinese-to-English Speech Translation. 201 On Path to Multimodal Generalist: General-Level and Gene...
-
[64]
Figure 125: Song Synthesis
多参考样例 generate song from the given description: An electronic that would loosely come under the electro-pop genre it is a song with vocals, a wobbly bass line, synths, white noises, and drums. Figure 125: Song Synthesis. 202 On Path to Multimodal Generalist: General-Level and ...
-
[66]
demo 导出时，需采用PDF格式，并且裁剪多于的空白部分，包括requirement
-
[67]
203 On Path to Multimodal Generalist: General-Level and General-Bench A.7.4 3D- RELATED TASKS Comprehension Tasks
多参考样例 generate music style from the given midi file and chord: Figure 126: Chord-based Music Style Transfer. 203 On Path to Multimodal Generalist: General-Level and General-Bench A.7.4 3D- RELATED TASKS Comprehension Tasks. Following, we showcase 13 3D-oriented comprehensive t...
-
[73]
Figure 136: 3D Normal Estimation
多多参考本样例 Estimate the normal for the given 3D point cloud. Figure 136: 3D Normal Estimation. 212 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: 3D Detection Paradigm: Comprehension Domain/Discipline: General-domain General/Universal Ca...
-
[79]
Figure 140: 3D-Point-Cloud Completion
多多参考本样例 Complete the given 3D point cloud. Figure 140: 3D-Point-Cloud Completion. 216 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Point Cloud to Mesh Reconstruction Paradigm: Generation Domain/Discipline: General-domain General/Uni...
-
[85]
Figure 141: Point-Cloud-to-Mesh Scene Reconstruction
多多参考本样例 Reconstruct the 3D mesh for the given 3D point cloud. Figure 141: Point-Cloud-to-Mesh Scene Reconstruction. 217 On Path to Multimodal Generalist: General-Level and General-Bench Skill (Meta-Task) Name: Text to Point Cloud Generation Paradigm: Generation Domain/Discipli...
-
[86]
不要随意调整：1）字体的大小，格式；2）整个图片区域的宽度
-
[87]
图片，视频的展示，必须清晰可见。不能随意压缩宽度、高度比例，使内容变形
-
[88]
图片放置位置可以在输入text的下方或者旁边，合理即可。
-
[89]
demo 导出时，需采用PDF格式（文件->导出->创建PDF）
-
[90]
使用工具（https://croppdf.com/zh/）裁剪多于的空白部分（包括requirements）， 提交裁剪后的PDF
-
[91]
I don't think I'll be getting anything,\
多参考样例 Passage: Oh man just read about Paul Walkers death. So young. Ugggh makes me sick especially when it’s caused by an accident. God bless his soul. – Jay Sean (@jaysean) December 1, 2013 Question: why is sean torn over the actor’s death? walker was young Figure 155: Tweet ...
2013
-
[92]
Designed data collection methods and se- lected models for verification
-
[93]
Responsible for all paper writing, illustra- tions, and polishing
-
[94]
Managed online deployment of data and au- tomated evaluation systems
-
[95]
7) Provided computing resources
Maintained the project website. 7) Provided computing resources. All audio-generation datasets All specialists in audio- generation tasks Audio MLLMs: WavLLM, ImageBind-LLM, Unified- io-2-XXL, ModaVerse-7b- v0, AudioGPT-GPT4, SpeechGPT-7B-com, LLaMA-Omni 2 Yuan Zhou Working fo...
-
[96]
Project co-leader: give the formal text and formula definitions for the 5 levels of the General-Level evaluation framework, along with the corresponding formula derivation
-
[97]
Led the image group, managing tasks and execution
-
[98]
Constructed and polished over 150 datasets; implemented around 30 SoTA specialists and 2 MLLMs
-
[99]
Verified task and data management, and de- ployed systems
-
[100]
Developed evaluation scripts and automated testing systems. All image-related datasets Image-oriented MLLMs: GPT4-o, GPT4-o-mini, GPT4-V 3 Juncheng Li Working for Image group Project co-leader for image group: Led the image group for supervised dataset collection and image-bas...
-
[101]
Project co-supervisor, conceptualized the idea of General-Level, and the entire process
-
[102]
/ / 32 Shuicheng Yan Project Supervision
Provided computing resources. / / 32 Shuicheng Yan Project Supervision
-
[103]
Project co-supervisor, co-conceptualized the idea of General-Level, and supervised the en- tire process
-
[104]
/ / 304 On Path to Multimodal Generalist: General-Level and General-Bench Table 119: List of some contributors without authorship
Provided computing resources. / / 304 On Path to Multimodal Generalist: General-Level and General-Bench Table 119: List of some contributors without authorship. # Name Contribution 1 Zhengzhe Liu Contributed to image group: assisted in image-oriented dataset preparation and mo...
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.